EEN:Edge Enhanced Network for Object Detection from SAR Images

Object detection in Synthetic Aperture Radar(SAR) imagery is becoming increasingly important due to its ability to function effectively in conditions where visible remote sensing is limited by weather factors. Despite these advantages, SAR images are frequently marred by coherent speckle noise and distinct edge features, which pose considerable challenges for object detection algorithms. We present an Edge Enhanced Network (EEN) designed to enhance the performance of SAR image object detection algorithms. The proposed network comprises two integral components: the Edge Enhancement Module (EEM) and the Context Fusion Module (CFM). The EEM applies specialized edge convolution to elements within SAR images, effectively amplifying edge features while reducing the influence of coherent speckle noise. The CFM primarily integrates contextual features, thereby enhancing multi-scale feature representation and bolstering the model's generalization capability. To validate the effectiveness of the EEN, this study conducts experiments using mainstream object detection networks, including YOLOX and RtmDet. Comparative and ablation experiments are carried out on the SSDD, RSDD-SAR, and SAR-AIRcraft datasets. The experimental results confirm the effectiveness of the EEN.