CFAR-Guided EfficientDet with Improved Loss Function for SAR Target Detection

Target detection has been a hot topic in the field of synthetic aperture radar (SAR) image analysis. In this paper, the one-stage object detector EfficientDet is applied in the SAR target detection. Since the labeled SAR data is limited and the SAR scene contains only a small number of target areas, the imbalance problem between positive and negative samples is serious in the network training process. To solve this problem, we introduce the Average-Precision Loss (AP-Loss) into the loss function of the class prediction net of EfficientDet. In addition, we integrate the binary indicator map obtained by the two-parameter constant false alarm rate (CFAR) detection into EfficientDet to guide the generation process of anchors. The experimental results on the miniSAR data demonstrate the effectiveness of the proposed method.