Joint Image Formation and Target Classification of SAR Images
| Autor: | Theresa Scarnati, Garrett Harris, Charles Connors |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Synthetic aperture radar
Image formation business.industry Computer science Fast Fourier transform ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Pattern recognition Iterative reconstruction Perceptron Discrete Fourier transform Statistical classification ComputingMethodologies_PATTERNRECOGNITION Computer Science::Graphics Computer Science::Computer Vision and Pattern Recognition Radar imaging Artificial intelligence business |
| Zdroj: | 2021 IEEE Radar Conference (RadarConf21). |
| DOI: | 10.1109/radarconf2147009.2021.9455246 |
| Popis: | Recently, a significant amount of research has gone into applying machine learning (ML) techniques to target classification from synthetic aperture radar (SAR) imagery. However, very little work has been done on using ML for SAR image formation. Classic image formation techniques, such as nonuniform fast Fourier transforms, backprojection, and sparsity based reconstruction, solve the SAR inverse imaging problem as a disjoint step in the overall task of target classification. In this work, we propose a novel joint image reconstruction and classification technique that solves both the SAR inverse imaging problem and the target classification task within a unified ML framework. Specifically, the proposed method optimizes for a SAR image through the use of a single layer perceptron (SLP) that maps two-dimensional discrete Fourier transform kernels to SAR sensor data. Through the mapping, the SAR image is output as the weights of the image formation network, allowing us to simultaneously update the image while training the classification network. Moreover, we adapt the ML algorithm to handle complex-valued SAR data, and thus allow for more accurate reconstructions. The proposed technique has the possibility to provide performance gains to target classification networks through implementation of this fully connected network. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|