Writing about aerospace and electronic systems, particularly with defense applications. Areas of interest include radar, sonar, space, satellites, unmanned plaforms, hypersonic platforms, and artificial intelligence.
Machine learning, or ML, is an offshoot of artificial intelligence
(AI) and has garnered significant attention due to the emergence of AI
tools like ChatGPT and DALL-E. It enables computer systems to adapt and
learn from experiences, making it a widely recognized concept. While its
popularity has grown recently, machine learning is already prevalent in
numerous real-life scenarios.
Curious about its everyday applications? Let's delve into some common examples of machine learning in action.
1. Personal Assistants and Chatbots
One of the most practical applications of machine learning is seen in
AI personal assistants and chatbots. Popular tools like Alexa, Google
Assistant, and Siri rely on ML algorithms that utilize Natural Language
Processing (NLP—what is NLP?)
and Deep Learning techniques to understand language patterns, tones,
and context. This enables them to engage in simulated conversations with
humans.
The ability to comprehend human language greatly simplifies our
interactions with computer systems. By providing a command or prompt to a
chatbot or AI personal assistant, it can accurately perform tasks or
provide relevant answers.
An example of ML in action is using chatbots in customer service.
Many e-commerce stores employ this feature, allowing customers to ask
questions and receive instant responses from the bots.
2. Email Autoresponders
Another common real-life application of machine learning is
ML-powered email autoresponders. You may have noticed that when you
receive an email in your Gmail account, it suggests accurate response
options based on the context of the conversation. This capability is
made possible through ML and NLP techniques.
Similarly, the emails you receive immediately after signing up for a
newsletter, making a purchase, or even abandoning a cart are all
automated. These emails are generated by software that utilizes such
models, enabling them to be sent out only when specific actions are
triggered. Additionally, this email autoresponder software ensures
customization and personalization in emails.
Using automated software for email responses offers several benefits, including:
Enhanced efficiency.
Improved customer service.
Personalized experiences by learning your preferences.
Cost savings as emails can be sent without human intervention.
3. Personal Finance and Banking
Machine learning algorithms have also entered the finance and banking sector, providing valuable applications. One of the ways banks use AI and ML algorithms
is advanced fraud detection techniques, which offer robust security for
your assets. ML models for fraud detection in banking can differentiate
between legal and illegal transactions by leveraging image and text
recognition methods to learn patterns and identify fraudulent
activities.
Machine learning is also beneficial in personal finance, particularly
in portfolio management. Online investment platforms, acting as
robo-advisors, utilize ML to assist in building, monitoring, and
automating the management of diversified portfolios. These platforms
learn about your preferences for specific assets or risks and help
construct your portfolio accordingly without human supervision.
Additionally, machine learning enables market forecasting in personal
finance. BL algorithms can predict stock prices and market trends by
analyzing historical data. This insight empowers you to develop
effective trading strategies and identify favorable trading
opportunities.
4. Healthcare and Medical Diagnosis
Machine learning has also emerged as a crucial tool in the healthcare
industry, offering numerous benefits for medical diagnosis, patient
care, and overall outcomes. It collaborates with various healthcare technologies that improve wellness in several ways.
Here are the six critical applications of machine learning in healthcare:
Machine learning algorithms analyze patient data, including
symptoms, medical records, lab results, and imaging scans, to assist in
precise disease diagnosis and prognosis.
By analyzing patient
characteristics, genetic information, treatment history, and clinical
data, machine learning develops personalized treatment plans tailored to
individual needs.
Machine learning facilitates the analysis
of X-rays, MRI scans, and pathology slides, automatically detecting
abnormalities, identifying specific features, and aiding radiologists in
disease diagnosis.
ML models optimize drug discovery
processes by enabling clinical trial optimization, patient recruitment,
and identifying suitable candidates for specific treatments.
Machine learning optimizes healthcare operations by providing supply
chain management systems, predicting equipment failure, and optimizing
resource allocation.
Machine learning enables predictive
analytics, working with Internet of Things (IoT) wearable devices to
monitor patients and provide early warnings.
These applications demonstrate the potential of machine learning to
revolutionize healthcare, improving diagnosis accuracy, treatment
efficacy, and overall patient care.
5. Self-Driving Cars
Machine learning plays a significant role in the development of
modern cars, with Tesla serving as a prominent example. Tesla's cars
rely on AI hardware provided by NVIDIA, incorporating unsupervised ML
models that enable self-learning object recognition and detection
capabilities. But it is not just Tesla with self-driving features.
These cars gather comprehensive information about their surroundings
and are equipped with various sensors such as cameras, LiDAR, radar, and
GPS. This data is then processed to ensure accurate perception and
effective decision-making. Self-driving cars utilize Simultaneous
Localization and Mapping (SLAM) techniques, leveraging sensor data to
create updated maps that aid navigation.
ML models further contribute to self-driving cars by determining
optimal paths and assisting in real-time decision-making. These models
also facilitate the development of adaptive systems capable of detecting
and predicting potential malfunctions in the vehicle.
By integrating machine learning, cars are becoming more intelligent,
autonomous, and capable of enhancing road safety and efficiency.
6. Commute and Transport
Machine learning algorithms have even elevated our commute and
transportation standards. Ride-hailing apps like Uber utilize ML models
to automate features such as ride pricing, pickup locations, optimal
routes, and estimated arrival time, making our daily commute more
convenient.
Google Maps is another valuable tool that leverages ML to enhance our
commute. Utilizing location data offers intelligent navigation, traffic
predictions, and personalized recommendations, ensuring efficient
travel.
In the aviation industry, ML algorithms enable airplane autopilot
systems, including commercial flights. This integration of AI and ML
ensures safe and reliable operations.
Furthermore, ML algorithms contribute to developing smart traffic
signal control systems. These systems analyze real-time traffic flow
data and adjust signal timings accordingly, reducing congestion and
waiting times to improve overall commute experiences.
Machine learning algorithms are also widely used in social networking
sites, where they have introduced numerous features to enhance the user
experience. Furthermore, many social media companies also use AI and ML to detect and prevent malicious attacks. Here are some examples:
Many popular social networking platforms such as Facebook,
Instagram, and Snapchat incorporate face recognition features for
applying various filters.
Social media sites utilize ML models to personalize posts based on individual user preferences.
Ads displayed on these platforms are tailored to users' interests, recommending relevant products and services.
Social networks provide suggestions for connections and friends based on users' existing networks.
Emotion analysis techniques are employed to analyze the sentiments conveyed by emojis.
These applications of machine learning algorithms in social
networking sites contribute to an improved user experience by providing
personalized content, relevant recommendations, and enhanced social
connections.
A Look Ahead at the Future of ML
The current trends in machine learning indicate its widespread
applicability across various industries, enabling system automation and
enhancing user experiences. From healthcare to finance, machine learning
transforms how we live, work, and interact with technology.
Looking ahead, the future of machine learning holds immense potential
for even more groundbreaking innovations. With the rise of AI-based
tools and software, machine learning algorithms are expected to continue
playing a crucial role. Their application extends to any domain that
requires big data analysis, pattern recognition, and AI implementation.
As machine learning advances, we can anticipate further advancements
and discoveries that will shape how we leverage this powerful technology
in diverse industries and aspects of our lives.
talk-technology.blogspot.com
In
recent years, the integration of robots and robotics in various
industries has been transforming the way we work and live. One area
where these advancements have significant implications is in law enforcement
and criminal justice. From enhancing crime prevention to improving
investigations and even assisting in dangerous operations, robots are
revolutionizing the field. This article explores the diverse
applications of robots and robotics in law enforcement, their benefits,
challenges, and the future they hold.
Crime Prevention and Surveillance:
Robots
equipped with advanced sensors and artificial intelligence (AI)
algorithms have proven to be valuable tools in crime prevention and
surveillance. Unmanned aerial vehicles (UAVs), commonly known as drones,
provide law enforcement agencies with aerial surveillance capabilities,
enabling them to monitor large areas efficiently and gather real-time
information. Drones are particularly useful in search and rescue
operations, crowd monitoring, and monitoring high-crime areas.
Additionally,
robotic surveillance systems with autonomous patrolling capabilities
can be deployed to monitor public spaces, detect suspicious activities,
and provide early warning to law enforcement officers. These systems
leverage technologies such as facial recognition, object detection, and
behavioral analysis algorithms to identify potential threats and enhance
public safety.\
Investigation and Forensics:
Robots
play a significant role in streamlining investigations and forensic
processes. Automated robots equipped with cameras and sensors can be
deployed to document crime scenes, collecting precise and reliable
evidence while minimizing human error. These robots can navigate
challenging terrains and access confined spaces, enabling investigators
to gather critical information without putting themselves at risk.
Moreover,
robots with sophisticated algorithms can analyze large volumes of
digital data, such as surveillance footage or text records, to assist in
identifying patterns, suspects, and connections. This AI-powered
analysis greatly speeds up investigations and provides valuable insights
to law enforcement agencies.
Bomb Disposal and Hazardous Situations:
One
of the most prominent applications of robots in law enforcement is bomb
disposal. Bomb disposal robots equipped with manipulator arms and
remote-controlled capabilities allow bomb technicians to approach and
disable explosive devices safely. These robots are equipped with
cameras, sensors, and tools that provide real-time feedback to the
operator, ensuring precise and controlled operations in high-risk
situations.
Additionally, robots can be used in hazardous material
incidents, such as chemical spills or nuclear incidents, to gather
information, assess risks, and even perform tasks like containment or
sample collection. By utilizing robots, law enforcement agencies can
safeguard the lives of their personnel and minimize the impact of
dangerous situations.
Rehabilitation and Support:
Robots
also play a role in the rehabilitation and support of offenders within
the criminal justice system. Robot-assisted therapy has shown promise in
helping individuals with behavioral disorders, substance abuse issues,
or post-traumatic stress disorder (PTSD) to cope and reintegrate into
society. These robots can provide structured interventions, monitor
progress, and offer a non-judgmental and consistent presence to support
individuals through their rehabilitation journey.
Ethical Considerations and Challenges:
While
the integration of robots in law enforcement brings numerous benefits,
it also raises important ethical considerations. Privacy concerns
related to surveillance technologies and the use of facial recognition
algorithms must be carefully addressed to balance public safety with
individual rights. Additionally, the potential for biased or
discriminatory algorithms requires close scrutiny to ensure fairness and
avoid perpetuating existing societal inequalities.
Moreover, the
reliance on robots should never undermine the value of human judgment
and empathy in law enforcement. While robots can enhance efficiency and
safety, human involvement and decision-making remain essential in
complex situations that require empathy, discretion, and ethical
judgment.
Conclusion:
The use of robots and
robotics in law enforcement and criminal justice systems is
revolutionizing the way crimes are prevented, investigations are
conducted, and public safety is maintained. From crime prevention and
surveillance to bomb disposal and rehabilitation
SkyFi lets you order up fresh satellite imagery in real time with a click
Aria Alamalhodaei
4–5 minutes
Commercial
Earth-observation companies collect an unprecedented volume of images
and data every single day, but purchasing even a single satellite image
can be cumbersome and time-intensive. SkyFi,
a two-year-old startup, is looking to change that with an app and API
that makes ordering a satellite image as easy as a click of a few
buttons on a smartphone or computer.
SkyFi doesn’t build or
operate satellites; instead, it partners with over a dozen companies to
deliver various kinds of satellite images — including optical, synthetic
aperture radar (SAR), and hyperspectral — directly to the customer via a
web and mobile app. A SkyFi user can task a satellite to capture a
specific image or choose from a library of previously captured images.
Some of SkyFi’s partners include public companies like Satellogic, as
well as newer startups like Umbra and Pixxel.
The startup is
taking a very 21st-century approach to the Earth observation industry.
SkyFi co-founders Bill Perkins and Luke Fischer emphasize that their
company is focused on user experience and creating a seamless purchasing
process for the consumer, contrasted sharply with what Fischer called
“business models based on the ’80s and ’90s.”
“We’re very
customer-focused,” Bill Perkins said on the TerraWatch Space podcast.
“The industry is science-focused and product-focused.”
SkyFi’s
mission has resonated with investors. The company closed a $7 million
seed round led by Balerion Space Ventures, with contributions from
existing investors J2 Ventures and Uber alumna’s VC firm Moving Capital.
Bill Perkins also participated. SkyFi has now raised over $17 million
to date.
The startup is targeting three types of customers:
individual consumers; large enterprise customers, from verticals
spanning agriculture, mining, finance, insurance and more; and U.S.
government and defense customers. SkyFi’s solution is appealing even
these latter customers, who may have plenty of experience working with
satellite companies already and could afford the high costs in the
traditional marketplace.
“Even though we have companies that are
multibillion dollar corporations using our platform that could afford to
have a multimillion dollar contract year with [any] public satellite
company, they’re being more cost conscious and that’s where this
offering of SkyFi comes in,” Fischer said.
Perkins and Fischer
experienced firsthand the pain points of the traditional satellite
imagery marketplace. For Perkins, the process of trying to buy satellite
images for his hedge fund was frustrating enough that he decided to try
to solve the problem himself.
He decided to team up with Fischer,
an Army aviation officer whose work experience includes stints at Uber
Elevate, Joby Aviation and Shield Capital. The two incorporated SkyFi in
December 2021 and officially launched the first product offering this
past January. As of today, the company has over 20,000 accounts from 185
countries registered on the platform.
One of their bets is that
the overly bureaucratic, time-intensive sales process has actually
constrained demand for satellite images. By making purchasing easier —
and providing transparent pricing — SkyFi anticipates whole new customer
bases and use cases opening up.
“There is no and will never be a ‘contact sales’ button on SkyFi,” Fischer said. “Because it just was ruining the industry.”
Looking
ahead, the Austin, Texas–based startup is planning on integrating
insight and analytics capabilities into the SkyFi app. This feature will
be especially useful for customers interested in hyperspectral or SAR
images. The company also plans to do more feature updates as it
integrates more providers — from satellites, to stratospheric balloons,
to drones — to the platform.
“I think of SkyFi as the Netflix of
the geospatial world, where I think of Umbra, Satellogic and Maxar as
the movie studios of the world,” Fischer said. “I just want them to
produce great content and put it on the platform.”
ICEYE has introduced a new imaging mode known as Dwell — with Dwell, ICEYE synthetic aperture radar (SAR) satellites remain focused on a specific point on the ground for an extended collection time.
This unique, long-duration, imaging mode allows ICEYE satellites to
capture significantly more information about the imaging target,
creating a powerful new capability for imagery users. For example, Dwell
can be used to determine the heading and speed of moving vehicles, and
under some circumstances, Dwell can even enable the discovery of
human-made objects that would otherwise remain hidden under tree canopy.
The Dwell product is unique in commercial remote sensing and will
transform how customers with demanding mission requirements exploit SAR
imagery.
The Dwell image product is created from a 25 second collection of the
imaging target, in contrast to the 10 second duration of a traditional,
high-resolution, ICEYE Spot image. This longer-duration collection
provides new information about activity on the ground while also
improving the image fidelity over a traditional SAR image. The result is
a powerful, new, imaging capability that will expand the use of SAR for
critical applications. Importantly, when combined with ICEYE’s
day/night all-weather imaging capability and tactical responsiveness,
Dwell provides an imagery capability that is superior to electro-optical
collections.
The Dwell product is a bundle with three components:
High-fidelity SAR image. A common feature in
high-resolution SAR images is speckle, which is the result of the
specific SAR collection geometry and gives SAR imagery a noise-like
appearance. The extended collection time for a Dwell image enables a
significantly larger synthetic aperture angle and produces pixels with
an along-track resolution of 5 cm. Standard processing combines pixels
together, a process called ‘multi-looking’, which significantly reduces
the speckle artifacts. The result is a high-fidelity image with superior
interpretability.
Color subaperture image. A Dwell collection is
particularly efficient at distinguishing human-made features, such as
vehicles and buildings, from natural backgrounds such as tree canopy and
vegetation. The color subaperture image preferentially distinguishes
human-made features, allowing analysts to quickly find objects and
features of interest among natural backgrounds.
Video. The video is built from the multiple
sub-images that comprise a Dwell collection. The video product enables
determination of vehicle heading and speed during the 25-second
collection, and provides additional context about activities and
patterns of behavior.
Dwell expands the powerful portfolio of high-tempo, day/night,
all-weather imaging capabilities enabled by the ICEYE constellation.
Today ICEYE offers a wide-area collection mode called Scan
for large area search; an intermediate, higher-resolution mode called
Strip for higher-resolution mapping and search, and highest-resolution
Spot and Spot Extended Area for high-confidence monitoring and mapping
applications. With the introduction of Dwell, users can now unlock new
insights about their areas of interest, doing so with capabilities
unique to the commercial SAR market.
“Dwell mode is ideal for customers who need to extract more
information from our radar imagery, including movement and direction. In
the long term, the research community, scientists, the SAR community,
and most importantly, our customers with demanding missions will benefit
from this latest ICEYE innovation. This product is another element in
using remote sensing to better understand and characterize what is
happening and changing on the ground in all lighting and weather
conditions.” — John Cartwright, Head of Data Product at ICEYE
New Space is based on a philosophy of creating less expensive
satellites in shorter periods of time, thanks to the falling costs and
miniaturisation of electronic parts. Nanosatellites and microsatellites
refer to miniaturized satellites in terms of size and weight, in the
range of 1-10 Kg and 10-100 kg, respectively. These are the fastest
growing segments in the satellite industry. With nanosatellites, the
benefits that were traditionally reserved exclusively for large
companies or space agencies with vast financial resources have been
democratised and are now accessible to companies of all types and
sizes. ‘CubeSat’ is one of the most popular types of miniaturized
satellites.
Small spacecraft, including nanosatellites, microsatellites, and
small satellites (smallsats), are an attractive alternative to
traditional, larger spacecraft due to reduced development costs,
decreased launch costs, and increased launch opportunities. CubeSats
reduce launch costs in two fundamental ways.
CubeSats were made possible by the ongoing miniaturization of
electronics, which allows instruments such as cameras to ride into orbit
at a fraction of the size of what was required at the beginning of the
space age in the 1960s. They don’t weigh that much, which means a rocket
doesn’t need a lot of fuel to heft them. In most cases, they also share
a rocket with a larger satellite, making it possible to get to space on
the coattails of the heavier payload.
CubeSat standardization opens up the possibility of using commercial
electronic parts and the choice of numerous technology suppliers, thereby
considerably cutting the costs of CubeSat engineering and development
projects in comparison with other types of satellites.
The trend toward small-sized spacecraft continues in government
applications and is even increasing in commercial space endeavors that
are funded by venture capital.
One of the major advantages of nano and microsatellites is reduced
delay and low cost of building and operating these satellites. This
means strongly reducing spacecraft lifecycle costs and lead time,
without reducing (and most likely increasing) performance. In turn, this
would allow the full potential of space to be exploited and space-based
systems to be competitive with ground-based systems that provide
similar services.
Modern small satellites has not only small size, light weight, high
technology, good performance, high reliability, short development cycle,
but also adaptability, ease of management, low risk, and thus has a
broad development and application prospects. It can be used as a single
satellite, but also satellites constellation.
Nanosatellites and microsatellites find application in scientific
research, communication, navigation and mapping, power, reconnaissance,
and others including Earth observation, biological experiments, and
remote sensing. There is also growing utilization of miniaturized
satellites for military and defense applications. Defense organizations
have been launching communication nanosatellites and microsatellites to
provide communication signals to soldiers stationed in remote locations
or in dense forests. The military needs more data bandwidth and
reliable communications infrastructure for its UAVs, which can be
fulfilled using constellations of nano and microsatellites.
Challenges
CubeSats can come in various sizes, but they are all based on the standard CubeSat unit, namely a cube-shaped structure measuring 10x10x10 centimetres with a mass of somewhere between 1 and 1.33 kg. This unit is known as 1U. After the first few years, this modular unit was multiplied and larger nanosatellites are now common (1.5U, 2U, 3U or 6U). Today, new configurations are under development.
To build such small, lightweight and intelligent spacecraft poses tremendous challenges.
Stringent monetary and mass/power/volume budgets.
A satellite payload is the main reason to launch the whole vehicle.
Thus, from a top-level point of view, the more ratio of payload mass to
total satellite mass (PM/TSM), the better. With ever-increasing progress
in microsatellite technology, PM/TSM as high as 10-25% is achievable,
at the present and in near future, respectively.
A significant disadvantage, however, of a small spacecraft is its
reduced or limited capabilities. The physical size of the small
spacecraft reduces the size of the payload and/or the number of payloads
that it can host, its propulsion capabilities, and its power.
The biggest challenge is miniaturization, virtually every spacecraft
subsystem needs breakthroughs in fully functional miniaturized
components in order to make the intelligent nanosatellite constellation
feasible.
Therefore there is a requirement to develop more capable payloads
supplied at reasonable prices. For instance, Surrey Satellite Technology
Ltd (SSTL) provides light-weight optical, navigation, and
communications payloads at exceptionally low prices
In-orbit autonomy
Highly-autonomous satellites are defined as those vehicles requiring
minimum contact with external sources (Terrestrial and/or Spaceborne) to
successfully accomplish their intended missions. Most microsatellites
are placed in LEOs, and communications gaps (time-intervals with no
contact opportunity) are inherent characteristics of LEOs. Thus,
logically, a given level of in-orbit autonomy must be accommodated
within the orbiting vehicle to perform mission-specific tasks, when out
of ground station visibility. Accommodation of a given level of onboard
autonomy is a sophisticated systems engineering activity confined by
inherent mass-/power-budget constraints of microsatellite missions and
also by LEO characteristics. For a microsatellite mission, once in
orbit, it is required to autonomously perform various self-management
and mission-specific tasks, to be utilized efficiently.
Attitude knowledge and control
The attitude Determination and Control System deals with the position
and orientation of the satellite in space, which is required for
maintaining stability and maneuvering for imaging and communications.
The accuracy and precision requirements are even more challenging for
small satellites where limited volume, mass, and power are available for
the attitude control system hardware. Lack of accurate three-axis
stabilized control capability has been a challenging obstacle in the
economical profitability of microsatellites. Today, three-axis control
with accuracies better than 1 degree are viable within microsatellite
stringent monetary and mass/power/volume budgets.
Attitude manoeuvrability
Attitude maneuverability is defined as the ability and agility of the
vehicle to align itself into a new desired orientation. Attitude
maneuverability has been traditionally one of the most demanding and
challenging in-orbit activities, possible only in complicated
several-hundred-kilogram satellites. However, recently, microsatellites
have proved their capability to accomplish demanding missions by
performing sophisticated Attitude maneuvers. Thus, now, microsatellites
can be scheduled to “look” in a certain direction, when over a desired
location. This capability gives the operators much more flexibility to
answer a user’s requests in a more rapid and responsive fashion
Communications
Applications of nanosatellites include data-intensive sensors like
hyperspectral video or imagers, which require higher data-rate downlink
capacity and greater power efficiency. Because nanosatellites have
restricted size, weight, and power (SWaP), this makes it challenging to
fit high-gain RF antennas for transmission to the ground. Hence, most
high-data-rate (i.e., higher frequency) nanosatellite missions require
high-gain ground stations with large dish diameters ranging from 5 m to
20 m and types of processing in the signal chain to RF signal issues and
attenuation.
In addition, it’s essential to ensure the converted signal data can
pass over the network stack to a host system in near real-time (in order
to analyze signal quality, errors, etc.). Moreover, obtaining RF
licenses with enough bandwidth for nanosatellite missions is
particularly difficult; so, it is becoming mandatory to manage multiple
narrow-bandwidth license requests. Thus, nanosatellites prefer SDRs over
older and less flexible RF communications.
Satellite constellations and clusters.
Small spacecraft are most commonly used in low Earth orbit, limiting
the number of observation opportunities for a particular area of the
Earth or space and the number of ground station downlink opportunities
for stored data. These constraints affect the complexity and types of
applications that small spacecraft can serve. Using multiple spacecraft
that work together can overcome many of these limitations and expand the
utility of small spacecraft. Two concepts for cooperative groups of
spacecraft are constellations and clusters.
A key technical challenge for small spacecraft, constellations, and
clusters is the communication of data. Communication challenges exist
for accommodating varying numbers of users, serving high user densities
in a given geographical area, and providing a consistent quality of
service for different types of applications (e.g., Internet access,
voice communication, machine-to-machine).
Radiation Susceptibility
The electronics are smaller and are therefore more sensitive to
radiation. Because they are small, they cannot carry large payloads with
them. Their low cost also means they are generally designed to last
only a few weeks, months or years before ceasing operations (and for
those in low Earth orbit, falling back into the atmosphere.)
Microsatellite technologies
Some of the technologies are integrated design of micro-nano
satellite technology are multiple system on chip, SoC, MEMS, 3D
multi-system integration technology to integrate the all functions into a
chip or multi-chip package. New technologies are being pioneered to
improve the use of CubeSats, such as a 2017 NASA parachute project that
could land the small satellites without the need of boosters.
Micro-nano satellite platform can be divided by function: on-board
computer, attitude and orbit control, monitoring and control, thermal
control, promotion, construction and power management seven aspects.
Miniaturization Technology
Miniaturization of space electronics requires an understanding of all
aspects of packaging technology, from the bare die level to the systems
level. Cost, performance, schedule, and reliability are the primary
drivers in the electromechanical development of space electronics.
Miniaturized sensors actuators
The small size of micro and nanosatellites constrains the small size
and mass requirements of Sensors, actuators, and other structural
elements needed for a specific mission. In all of these systems the
minimization of power facilitates overall mass reduction and is an
important design parameter.
The new tiny optical imager developed by Spain-based start-up
SATLANTIS can take images with a resolution of less than a metre – a
remarkable technological achievement for an under 80 kg-class
microsatellite. Their compact size, low mass (less than 15 kg), short
delivery times (5 times faster) and decreased production costs (5 times
more affordable) give it a powerful competitive edge over the latest
tailor-made imaging systems on the market,” explains project coordinator
MarÃa DasÃ.
Resolution (in remote-sensing systems)
Remote sensing applications are among the most promising applications
of LEO microsatellites. LEO is the main domain of microsatellite
missions. This has been due to low launch costs and the limited
capabilities of microsatellites. The main requirement of such systems
comes in the form of spatial resolution or GSD (Ground Sample Distance).
Under 1-meter resolution radar imagery achieved by ICEYE
CEYE, a small satellite synthetic-aperture radar (SAR) technology
company, achieved better than 1 meter resolution imagery from under-220
pound SAR satellites. ICEYE’s newly deployed Spotlight imaging mode
enables under 1 meter radar imaging from the company’s satellites. With
Spotlight imaging, the satellite focuses its energy on a smaller area
for a longer time, resulting in more data received from the same
location. This in turn can be processed into more detailed imagery. Very
high-resolution radar satellite images are helpful for both
distinguishing small objects, and for accurately classifying larger
objects such as vessels. These added capabilities of ICEYE’s SAR
satellites are valuable in resolving challenges in sectors such as
emergency response, finance, civil government, and maritime security
The optimization of navigation system.
Application of MEMS technology can be effectively optimized for
navigation guidance sensors, such as gyroscope and accelerometers, MEMS
technology designed gyroscope area has been reduced to a few
centimeters, monolithic integrated three-axis gyroscope gradually
realized, and significantly reduce the satellite navigation sensor size
and power consumption. MEMS systems can replace some of the existing
machinery and equipment, to achieve the sensing of environmental and
location. Therefore, the design of micro-nano satellites can be used the
Local MEMS devices, for laying the foundation for an multi-system
integrated assembly in the future.
Microsat Power Systems and the IPS
Traditional spacecraft power systems that typically require some
level of ground management incorporate a solar array energy source, an
energy storage element (battery), and battery charge control and bus
voltage regulation electronics to provide continuous electrical power
for spacecraft systems and instruments. The cost-effective operation of a
microsat constellation requires a fault-tolerant spacecraft
architecture that minimizes the need for ground station intervention by
permitting autonomous reconfiguration in response to unexpected fault
conditions. The microsat power system architecture provides unregulated
voltages that can be distributed to spacecraft systems and instruments
in a traditional manner or used to power dedicated spacecraft loads
through linear regulators or power converters.
Centralized computer unit.
The use of centralized data processing mode, will merge computing
resources ,such as attitude, orbit control, on-board management and
ambient temperature, using a single high-performance computing unit for
centralized management. Current computing requirements can be satisfied
by 100MIPS processor, embedding more than 1M memory, which fully meet
the data processing requirements of micro-nano or small satellites.
Centralized data management to reduce the complexity of the system,
redundant design increases the system reliability.
SoPC Flight Control Design and Implementation
On-chip multi-system integration technology as a technology platform
has now completed a chip for the function of control, data processing
and navigation, mainly for the application of small satellites and
onboard flight control. Design objectives and requirements: 100MIPS
computing power, storage space is greater than 10M, programmable space,
I2 bus, satellite navigation, inertial platform interface, high speed
digital to analog conversion. The general requirements for flight
control, there must be a high-performance computing power and flexible
software and hardware programming capabilities, navigation and
positioning capabilities.
Advanced Antennas
Antennas are key components that enable small satellites to receive
and transmit electromagnetic signals. Onboard small satellites, there
are a number of antennas for different functions. Due to the limited
volume onboard small satellites, it is important to optimize the antenna
designs, which directly determine the performance of all wireless
systems onboard satellites, such as telemetry, tracking, and control
(TTC), high-speed data downlink, navigation, intersatellite
communications, intrasatellite communications, wireless power transfer,
radars and sensors, etc.
Many antenna types with different operating frequency bands are
proposed for CubeSat applications. Some of the antenna designs include
patch antennas, slot antennas, dipole and monopole antennas, reflector
antennas, reflectarray antennas, helical antennas, metasurface
antennas and 3 millimeter and sub-millimeter wave antennas. In
addition, antennas are also classified according to their operating
frequency bands, e.g., VHF, UHF, L, S, C, X, Ku, K/Ka, W and mm/sub-mm
wave bands.
Communications using Software Defined Radio (SDR)
The Microsatellites also carry a radiation-hardened SDR transponder
that leverages existing designs and enhance capabilities in the
commercial sector. A Software Defined Radio (SDR) concept uses a minimum
amount of analog/radio frequency components to up/downconvert the RF
signal to/from a digital format. Rest all other processing (filtering,
modulation, demodulation, etc.) is done in digital domain in software.
This allows Software to be reused across products, reducing software/
hardware costs dramatically. As new telecommunication technologies
emerge, incorporating them into the SDR fabric will be easily
accomplished with little or no requirements for new hardware. New
features and capabilities, such as encoding and decoding algorithms,
filters, and bit synchronizers, can be added to the existing
infrastructure without requiring major new capital expenditures,
allowing implementation of advanced features in the communication
systems.
Internet of Things (IoT) start-up Fleet Space Technologies has signed
a contract to launch two of its satellites onboard the next Rocket Lab
flight, dubbed “It’s Business Time.” The launch is scheduled to occur in
November 2018 from New Zealand. Proxima I and II were both designed and
built by Fleet. The two satellites will mark the first commercial tests
of the company’s software-defined radios (SDRs), enabling the company
to move data in both S-band and L-band frequencies in space. SDR is a
key technology being used by a number of small satellite start-ups.
Optical Communication on CubeSats
The purpose of a telecommunication subsystem for a microsatellite is
to serve as a communication link between the microsatellite and the
ground station. Currently, CubeSats rely completely on radiofrequency
communications using amateur low-speed UHF systems (with omnidirectional
dipole antennas) due to its availability and lower cost, with rates in
the order of kbit/s or tens of kbit/s from Low-Earth Orbit (LEO). As a
result, the potential of most CubeSat missions is being limited by their
communication capabilities, therefore there is a tendency to move to
higher frequencies, especially to X-band, where more bandwidth is
available, as an alternative to achieve higher data rates with smaller
ground antennas. The Earth-imaging company Planet, has shown a
remarkable success in these X-band communication systems, using COTS
components and 5-m class ground antennas to achieve sustained data rates
in the order of 100-200 Mbit/s with the latest generation of their
‘Dove’ CubeSats.
“Small satellites can’t use these bands, because it requires clearing
a lot of regulatory hurdles, and allocation typically goes to big
players like huge geostationary satellites,” said Cahoy, who also has an
appointment in MIT’s Department of Earth, Atmospheric and Planetary
Sciences. Furthermore, the transmitters required for high-rate data
downlinks can use more power than miniature satellites can accommodate
while still supporting a payload. For these reasons, researchers have
looked to lasers as an alternative form of communication for CubeSats,
as they are significantly more compact in size and are more power
efficient, packing much more data in their tightly focused beams.
Space optical communications could play an important role by
improving of several orders of magnitude in the transmission capacity,
while keeping a low size, weight and power, thereby enhancing the
potential of CubeSats with growing bandwidth requirements. The key
components of a CubeSat lasercom system are the optical-power generation
and the pointing capability. Generating several Watts of optical power
imposes requirements that go beyond what a CubeSat can usually deliver,
although an effective strategy to alleviate this is to include secondary
batteries in the power management system.
FPGA’s & ASICS
At the chip level, custom and semicustom ICs and field-programmable
gate arrays can integrate many functions onto the fewest number of ICs
to reduce the parts count in the implementation of the electronics.
Fewer components enhance reliability as well as reduce weight and
volume. Among other important aspects in the design and development of
application-specific ICs, besides cost and schedule concerns, are the
radiation tolerance of the design and its implementation on the selected
IC foundry line.
Multi-system assembly techniques
Multi-system assembly techniques, composed of a sensor unit, a data
processing unit and a FPGA (implement diversity peripheral interface or
bus), use 3D packaging technology to achieve integration. Multiple chips
of the electronic systems are integrated or assembled in a single
package, so that you can effectively achieve minimum micro-nano
satellites. Multi-system technology assembly technology uses of
micro-nano assembly technology, system in package(SiP) technology and
other assembly techniques, completing multi-system interconnect and
communicate.
Integrated electronics and Architecture
This includes the use of highly integrated electronics technologies
as well as the integration of functions classically implemented in
separate elements with the IPS, which combines energy storage, solar
array electronics, and charge control electronics into a single
structural element. With a compact electronics architecture, the
electronics must be designed to incorporate the fewest components on
boards that can be integrated in the most compact format possible.
At the component level, parts selection focuses on the integration of
the smallest outline packaged parts (i.e., chip scale packaging) that
are available for space use. Indeed, the smallest area required for the
implementation of electronics involves mounting just the IC die itself
and dispensing with the housing package for the die. The overall
architecture is also important, with a conventional satellite
electronics architecture, functional electronics subsystems are
connected with wiring harnesses distributed throughout the satellite,
adding considerable weight and volume. Typically, the harness is about
7% of the dry spacecraft weight.
With an integrated electronics module (IEM) approach, however, much
of the onboard electronics are packaged onto electronics boards that are
integrated into a single housing in which the boards are connected in a
backplane fashion. This can result in considerable weight and volume
savings. Standardized microsatellite platform helps the system
integration and miniaturization.
Low power electronics
For power savings, the design of low-power electronics factors in
power management techniques and considers low supply voltages. The power
dissipated in CMOS integrated circuits (ICs) is a function of
C*sqr(V)*f, where C is the load capacitance, V is the supply voltage,
and f is the clock frequency. Research in the development of ultra-low
power <1 V CMOS ICs suitable for space is encouraging and will likely
result in major reductions in power dissipation in digital electronics.
A study showed that the use of ultra-low voltage IC technology can
yield a 20 to 50% saving in power, largely from a reduction of digital
electronics power consumption.
NSST is a discrete form of Shearlet transform, and it
differs from other multi-scale transformations by avoiding up-down
samplers. NSST consists of two main stages, multi-scale and
multi-directional separations.
Multi-scale
The Laplacian Pyramid (NSLP) without subsampling produces low and
high frequency images whose size is the same as the size of the source
image.
Multi-directional
Versatility is achieved by using different "combinations of Shear Filters" in the so-called polar (pseude-polar) coordinate
NSST Steps
The process steps performed to obtain the NSST coefficients of an image of NxN size at a fixed resolution scale are as follows:
Laplacian pyramid is applied to the image. Low and High pass sub-images are obtained.
The fourier transformations of the high pass sub-images are calculated and transformed into the Polar coordinate system.
Bandpass filter is applied to Polar coordinate system
transformations and Fourier transforms (FFT) of Shearlet coefficients
are obtained.
The Inverse Fourier Transform (IFT) is applied to obtain the
Shearlet coefficients and the transformation is performed to the
Cartesian coordinate system.
Pipeline
Firstly, the input image must be converted to the Intensity channel.
At this stage, only Y channel will be used for NSST. You can use ConvertBMPToIntensity() for this process. General formula :
Y = (0.11 * Red + 0.59 * Green + 0.3 * Blue)
Intensity image given to NSST function. Low (AFK) and High (YFK)
Frequency coefficients are obtained at its output. You can use NsstDec1e() for this process.
You can get the input Intensity image using TNSST (Inverse NSST). You can use NsstRec1() for this process.
summary of "An NSST-Based Fusion Method for Airborne Dual-Frequency, High-Spatial-Resolution SAR Images" https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10109765
The paper titled "An NSST-Based Fusion Method for Airborne Dual-Frequency, High-Spatial-Resolution SAR Images" presents a fusion technique for enhancing the quality of airborne synthetic aperture radar (SAR) images. The method utilizes the Non-Subsampled Shearlet Transform (NSST) to combine images acquired at different frequencies and improve their spatial resolution.
The authors begin by explaining the limitations of traditional SAR imaging techniques, which often suffer from low spatial resolution due to the system constraints. To overcome this limitation, the proposed method exploits the complementary information provided by dual-frequency SAR images.
The fusion process consists of several steps. First, the dual-frequency SAR images are decomposed into different frequency subbands using the NSST. The NSST is chosen because it effectively captures the directional and scale information in images. This decomposition leads to a set of subband coefficients at multiple scales and orientations.
Next, a fusion rule is applied to the NSST coefficients to combine the information from both frequency bands. The fusion rule is based on the principle of selecting coefficients that possess high-energy content while preserving the fine details. This step ensures that the fused image maintains both the high-frequency information and the spatial details from the original images.
After the fusion step, the fused NSST coefficients are reconstructed to obtain the final fused SAR image. The reconstruction process involves inversely transforming the NSST coefficients in each subband to their corresponding spatial domain using the inverse NSST.
To evaluate the performance of the proposed fusion method, the authors conducted experiments using real airborne dual-frequency SAR data. They compared the fused images with the original dual-frequency images as well as other existing fusion techniques. The evaluation metrics included visual quality assessment and quantitative measures such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM).
The experimental results demonstrate that the NSST-based fusion method effectively enhances the spatial resolution of airborne dual-frequency SAR images while preserving important details. The fused images exhibit improved visual quality and outperform other fusion techniques in terms of PSNR and SSIM.
In conclusion, the paper presents a detailed description of an NSST-based fusion method for enhancing the spatial resolution of airborne dual-frequency SAR images. The method utilizes the NSST decomposition and fusion rules to effectively combine the complementary information from different frequency bands. The experimental results validate the effectiveness of the proposed method in improving the quality of SAR images for various applications.
The Non-Subsampled Shearlet Transform (NSST) is a mathematical tool used for image analysis and processing. It is an extension of the traditional Discrete Wavelet Transform (DWT) and provides a more flexible and directional representation of images.
The NSST is designed to capture the sparse and directional characteristics of images by using shearlets, which are specialized waveforms that are elongated and oriented in different directions. Shearlets are well-suited for representing edges, curves, and other geometric features in images.
Compared to the DWT, which only captures scale information, the NSST captures both scale and directional information. This makes it particularly useful for tasks such as image denoising, image fusion, and feature extraction.
The NSST operates in a multiscale and multidirectional manner. It decomposes an input image into a set of subbands at different scales and orientations. This decomposition is achieved by convolving the image with a set of shearlet filters. Each subband represents a specific range of frequencies and directions.
The NSST coefficients obtained from the decomposition can be used for various image processing tasks. For example, in image fusion, NSST coefficients from different images can be combined to create a fused image that contains the most salient information from each input image.
The NSST reconstruction process involves taking the inverse transform of the NSST coefficients to reconstruct the image in the spatial domain. This allows for the recovery of the original image from its shearlet representation.
Overall, the NSST provides a powerful tool for analyzing and processing images, especially in scenarios where capturing both scale and directional information is important. Its ability to represent image features in a sparse and localized manner makes it particularly useful for applications in image fusion, denoising, texture analysis, and other image processing tasks.
The non-subsampled shearlet transform (NSST) is a multiscale and directional transform that is well-suited for representing a wide variety of signals, including images, audio, and video. NSST is a discrete form of the shearlet transform, which is a continuous-time transform that was introduced by Mallat and Hwang in 1999.
NSST is a non-subsampled transform, which means that it does not involve any upsampling or downsampling of the input signal. This makes NSST more efficient than other multiscale transforms, such as the wavelet transform, which do involve upsampling and downsampling.
NSST is also a directional transform, which means that it can represent signals that have directional features. This makes NSST well-suited for representing images, which often have edges and other directional features.
NSST has been used for a variety of signal processing applications, including image denoising, image compression, and image segmentation. NSST has also been used for audio processing applications, such as audio denoising and audio compression.
Here are some of the advantages of using NSST:
It is a non-subsampled transform, which makes it more efficient than other multiscale transforms.
It is a directional transform, which makes it well-suited for representing signals that have directional features.
It has been shown to be effective for a variety of signal processing applications.
Here are some of the disadvantages of using NSST:
It is a relatively new transform, so there is less research on it than on other transforms, such as the wavelet transform.
It can be computationally more expensive than other transforms, such as the wavelet transform.
Overall, NSST is a powerful and versatile transform that can be used for a variety of signal processing applications.
How to Turn Your Voice to Text in Real Time With Whisper Desktop
Odysseas Kourafalos
9–11 minutes
The very same people behind ChatGPT have created another AI-based
tool you can use today to boost your productivity. We're referring to
Whisper, a voice-to-text solution that eclipsed all similar solutions
that came before it.
You can use Whisper in your programs or the command line. And yet,
that defeats its very purpose: typing without a keyboard. If you need to
type to use it, why use it to avoid typing? Thankfully, you can now use
Whisper through a desktop GUI. Even better, it can also transcribe your
voice almost in real time. Let's see how you can type with your voice
using Whisper Desktop.
What Is OpenAI's Whisper?
OpenAI's Whisper is an Automatic Speech Recognition system (ASR for
short) or, to put it simply, is a solution for converting spoken
language into text.
However, unlike older dictation and transcription systems, Whisper is
an AI solution trained on over 680,000 hours of speech in various
languages. Whisper offers unparalleled accuracy and, quite impressively,
not only is it multilingual, but it can also translate between
languages.
More importantly, it's free and available as open source. Thanks to
that, many developers have forked its code into their own projects or
created apps that rely on it, like Whisper Desktop.
Despite its official-sounding name, Whisper Desktop is a third-party
GUI for Whisper, made for everyone who'd prefer to click buttons instead
of typing commands.
Whisper Desktop is a standalone solution that doesn't rely on an
existing Whisper installation. As a bonus, it uses an alternative,
optimized version of Whisper, so it should perform better than the
standalone version.
You're on the other end of the spectrum, and instead of seeking an
easier way to use Whisper than the terminal you're seeking ways to
implement it in your own solutions? Rejoice, for OpenAI has opened access to ChatGPT and Whisper APIs.
Download & Install Whisper Desktop
Although Whisper Desktop is easier to use than the standalone
Whisper, its installation is more convoluted than repeatedly clicking
Next in a wizard.
Under Assets, click WhisperDesktop.zip and download it to your PC.
Extract the downloaded archive to a folder and use your file manager to
visit it. Inside you will find the Whisper Desktop application.
Double-click on it to run it.
You also need a Whisper language model in GCML
binary format. Whisper Desktop will provide you with two links for
acquiring one. Skip the second link for generating your own model since
it's a more complicated process. Click on Hugging Face to open that page in your default browser, from where you can download a ready-to-use file.
The version of Whisper Desktop we used while writing this article
provided a link to an obsolete repository at Hugging Face. If you meet
the same problem, notice a link to a new location. Click on it to visit the new repository.
Click on the link that will take you to the available models.
From that list, click on either the ggml-medium.bin or ggml-medium.en.bin, depending on if you want multilingual or English-only support in Whisper.
Finally, you should have reached your destination. Notice the line
stating that this file is stored with Git LFS and is too big to display,
but you can still download it. Click on download to do precisely that.
When the file completes downloading, use your favorite file manager
(File Explorer will do) to move the downloaded language model file into
the same folder as Whisper Desktop.
Transcribing With Whisper Desktop
Transcribing with Whisper Desktop is easy, but you may still need one or two clicks to use the app.
Rerun Whisper Desktop. Does it (still) miss the correct path to your downloaded language model? Click on the button with the three dots on the right of the field and manually select the file you downloaded from Hugging Face.
From this spot, you can also use the drop-down menu next to Model Implementation to choose if you want to run Whisper on your GPU (GPU), on both the CPU and GPU (Hybrid), or only on the CPU (Reference).
The Advanced button leads to more options that
affect how Whisper will run on your hardware. However, since the button
clearly states they are advanced, we suggest you only tweak them if you
are troubleshooting or know what you are doing. Setting the wrong
options values here can impose a performance penalty or render the app
unusable.
Click on OK to move to the app's main interface.
If you already have a recording of your voice you want to turn into written text, click on Transcribe File and select it. Still, we will use Whisper Desktop for live transcription for this article.
The options offered are straightforward. You can select the language Whisper will use, choose if you want to translate between languages and enable the app's Debug Console.
Most English-speaking users can safely skip those options and only
ensure the correct audio input is selected from the pull-down menu next
to Capture Device.
Make sure Save to text file and Append to that file are enabled to have Whisper Desktop save its output to a file without overwriting its content. Use the button with the three dots on the right of the file's path field to define said text file.
Click on Capture to begin transcribing your speech to text.
Whisper Desktop will show you three indicators for when it detects
voice activity, when it's actively transcribing, and when the process is
stalled.
You can keep talking for as long as you like, and you should
occasionally see the two first indicators flashing while the app turns
your voice into text. Click Stop when done.
The text file you selected should open in your default text editor,
containing in written form everything you said until you clicked Stop.
We should note that you can also do the opposite of what we saw here:
convert any text to speech. This way you can listen to anything as if
it were a podcast instead of tiring your eyes squinting at screens. For
more info on that, check our article on some of the best free online tools to download text-to-speech as MP3 audio.
Whisper Desktop Voice-Typing Tips
Although Whisper Desktop can be a lifesaver, enabling you to write
with your voice much quicker than you could type, it's far from perfect.
During our testing, we found that it may occasionally stutter, skip
some words, fail to transcribe until you manually stop and restart the
process, or get stuck in a loop and keep re-transcribing the same phrase
repeatedly.
We believe those are temporary glitches that will be fixed since the standalone Whisper doesn't exhibit the same issues.
Apart from those minor bumps, turning your voice to text should be
effortless with Whisper Desktop. Still, during our tests, we found that
it can perform even better if...
Instead of uttering only two or three words and then pausing,
Whisper can understand you better if you go on longer. Try to at least
give it an entire sentence at a time.
For the same reason, avoid repeatedly starting and stopping the transcription process.
Whenever you realize you made a mistake, ignore it and keep going.
Loading and unloading the language model seems to be the most
time-consuming part of the process with the current state of Whisper and
our available hardware. So, it's quicker to keep talking and then edit
out your mistakes afterward.
As with the standalone version
of Whisper, it's best to use the optimal language model for your
available hardware. You can use up to the medium model
if your GPU has 8GB of VRAM. For less VRAM, go for the smaller models.
Only choose the slightly more accurate but also much more demanding large model if you use a GPU with 16GB of VRAM or more.
Remember that the larger the language model, the slower the
transcription process. Don't go for a model larger than needed. You'll
probably find Whisper Desktop can already "understand you" most of the
time with the medium or smaller models, with only one or two errors per
paragraph.
Are You Still Typing? Use Your Voice With Whisper
Despite requiring some time to set up, as you will see when you try
it, Whisper Desktop performs much better than most alternatives, with
much higher accuracy and better speed.
After you start using it to type with your voice, your keyboard may look like a relic from ancient times long gone.
