Electro-Optic Infrared Image Outlier Detection
EO/IR image outlier detection is a process that enables persistent observation systems a reliable way to flag image frames that might show the presence of a target or collection of features. Image outlier algorithms are considered an established but evolving technology that benefits from large volumes of EO/IR imagery produced under known, or, better yet, controlled conditions. ThermoAnalytics’ combination of software developers, modeling experts and scientists collaborate to produce virtual datasets for training such as deep learning outlier classifiers.
Wide Range of Data Renderings
We produce accurate synthetic EO/IR targets and backgrounds across the full gamut of operational, situational, and climate/weather profiles. These datasets can be used to teach learning algorithms. By combining variations in activity, weather, time of day, global location, even clothing or optical properties, our process can yield a wider range of training data renderings than is possible through physical acquisition in the field. Using atmospheric modeling, we can also include the effects of atmospheric transmission of the source radiance and determine what a sensor may detect at-range, with the source radiance attenuated.
Deep Neural Network Focus
Our image outlier data sets include both normal (target not present) and abnormal (target-present) renderings in ratios that enable a balanced neural network to be developed. While neural networks trained on generic datasets could be used with the hope of detecting targets, they would be less accurate than a properly trained deep neural network focused on the categories of targets your system would require.
Simplifying Data Set Training
In operation, these image outliers would then be evaluated by a human for secondary detection analysis, having been previously evaluated against a deep learning model, trained by our comprehensive library of virtual images. Data sets could also be trained for specific operational locations or mission profiles, rather than developing a universal data set that includes all global regions and background types.
You Might Also Be Interested In:
Industry-Leading Signature Management
MuSES (Multi-Service Electro-Optic Signature) is an industry-leading signature management software that provides complete thermal modeling and infrared signature prediction capabilities. MuSES allows you to simulate any environment or condition and generate accurate signature predictions. It provides many powerful features that allow you to model plumes, wakes, humans, atmospheric effects, and much more.
Predicting Survival + Effectiveness
Predicting human thermal survival involves simulation across several distinct domains to be computed simultaneously – the environmental conditions over time, a detailed human physiology model that includes thermoregulation, and clothing models to capture both heat transfer and mass transfer resistance. Humans are thermally quite fragile and many critical biological processes are highly sensitive to temperature, including cognition. Thermal survival is limited to the study of how the body responds to dehydration, hypo- and hyperthermia, ignoring the situational effects of reduced cognitive capacity on the choices an individual makes when overheated or suffering from cold exposure.
Adversary signature prediction
EO/IR image outlier detection is a process that enables persistent observation systems a reliable way to flag image frames that might show the presence of a target or collection of features. Image outlier algorithms are considered an established but evolving technology that benefits from large volumes of EO/IR imagery produced under known, or, better yet, controlled conditions. ThermoAnalytics’ combination of software developers, modeling experts and scientists collaborate to produce virtual datasets for training such deep learning outlier classifiers.
Effective Strategies for Signature Analysis
Signature requirements for EO/IR seek to define how easily a vehicle, ship, aircraft, or spacecraft can be “acquired” or recognized as distinct from the background environment. It is similar to defining a maximum signal-to-noise ratio for a given platform: the lower the signal, the less distinct it is from the noisy background imagery. Because the requirements for signature often dictate the maximum contrast relative to the background, leading OEMs contract with ThermoAnalytics to plan ahead for effective strategies to minimize or control signatures, leading to improved performance and what is often termed, “low observables” design.