Enhancing SAR Automatic Target Recognition with Advanced Target Generative Adversarial Network

These are the novel approaches taken by us to minimize the hurdles ATR poses for SAR systems. Thus, we designed ATGAN-an Advanced Target Generative Adversarial Network with a dual-network architecture toward image generation realistic for SAR target generation. Substantial experimentations have improved the recognition accuracy and robustness as well, compared with traditional models like SVM, KNN and LSTM networks. The performance metrics of ATGAN are truly phenomenal; the sensitivity score was $0.9998 \%$, specificity at $0.9997 \%$, and an accuracy of $0.9999 \%$. Its F1 value had reached an incredible value of $\mathbf{0. 9 9 9 6 \%}$ to show its excellent balance between precision and recall. The SVM, KNN, and LSTM models scored perfectly well but were a couple of notches behind in the critical parameters of ATGAN. Results show that the ATGAN-based approach outperformed methods reported in literature for SAR ATR tasks, especially in complex and dynamic environments where accuracy is of high priority.