Multi-View Fusion SAR Target Recognition Method with LSTM Temporal Decision-Making

This paper proposes a sequential multi-view SAR image target recognition method based on LSTM temporal decision-making, aiming to address the misclassification issues caused by traditional single-frame classification algorithms neglecting temporal correlations in dynamic monitoring scenarios. By integrating the nonlinear feature extraction capability of Convolutional Neural Networks (CNN), the margin maximization advantage of Support Vector Machines (SVM), and the temporal modeling ability of Long Short-Term Memory (LSTM) networks, we achieve multi-level information fusion from spatial features to temporal dependencies. Experimental results demonstrate that the sequential recognition algorithm incorporating LSTM improves recognition accuracy by $10 \%-20 \%$ compared to the traditional CNN+SVM method in SAR target identification. Analysis of consecutive-view sample groups verifies that LSTM effectively corrects misclassifications caused by local feature perturbations (e.g., rotation, noise) through temporal correlations. Otherwise, additive Gaussian noise is introduced to simulate different signal-to-noise ratio (SNR) environments, validating the algorithm’s robustness under complex noise conditions.