Aerial Vehicles Automated Target Recognition of Synthetic SAR Imagery Using Hybrid Stacked Denoising Autoencoders
Synthetic Aperture Radar (SAR) technology offers innovative remote sensing opportunity for the area of surveillance applications. However, for the Automatic Target Recognition (ATR) from SAR data, there is a need for large-scale multi-look imagery of the target objects of interest (TOI’s) from different perspective viewing angles - that is rarely available publically specially for the aerial vehicles. Such large-scale datasets can be very instrumental for the training of deep learning classifiers as well as for achievement of more robust transfer learning. We address this shortcoming by introducing IRIS Electromagnetic (EM) modeling and simulation system for virtual staging and automatic generation of realistic synthetic (i.e. simulated) multi-look SAR imagery of aerial vehicles for the purpose of training of the ATR systems. Primarily, a collection of 250 physics-based CAD models containing different aerial and ground vehicle objects was created prior to obtaining their simulated EM reflectivity maps via an optimized ray tracing technique. Furthermore, to condition our synthetically generated reflectivity map imagery with characteristics pertaining to the environment clutters like ground, grass, and asphalt, we introduced different noise models to represent these backgrounds radiation backscattering. Moreover, we introduced a technique for realistic modeling of shadows pertaining to the TOI’s SAR imagery under different azimuth and elevation scanning perspectives. Furthermore, we developed a robust image processing technique to further improve the fidelity of the generated synthetic imagery to reflect the SAR speckle noises off of the high reflectivity points in the scene. To test and verify the dependability of this proposed approach, we compared our simulated SAR imagery results against a number of comparable military and commercial vehicles from the real MSTAR dataset. Lastly, we implemented a Hybrid Stacked Denoising Auto-Encoder (HSDAE) as an effective denoiser and classifier model. Using the simulated ground-truth object reflectivity maps of test aerial vehicles, we trained our model to perform the denoising and classification tasks. This model performs two concurrent operations – denoising and classification. Through the denoising operation, the test objects’ shape, size, and orientation attributes are recomposed. Then, in the next step, the recomposed imagery results are employed for performing the object classification. Comparing this system’s performance against those from different CNN models shows higher due to the zero noise present in the image after applying the SDAE.
Mechanical engineering|Computer science|Electrical engineering|Artificial intelligence
"Aerial Vehicles Automated Target Recognition of Synthetic SAR Imagery Using Hybrid Stacked Denoising Autoencoders"
ETD Collection for Tennessee State University.