Improved YOLOv8 SAR Image Aircraft Object Detection Method

Due to the poor model detection accuracy, easy overfitting and weak generalization in aircraft target detection under complex background in SAR images, this thesis introduces an enhanced SAR aircraft object detection algorithm, YOLOv8-GD, which is based on the YOLOv8 framework. The Gather-and-Distribute (GD) machine were used to enhance the capability of multi-scale feature fusion and improve the correct prediction rate. The GSConv is employed in the backbone network, and Wise-IoU (WIoU) loss function are used to raise the detection accuracy of the target and refine the model's generalizability. Experiments were carried out on the SAR-Aircraft-1.0 dataset. In comparison to the current mainstream methods, the accuracy of detecting Aircraft targets in SAR images was effectively improved, which reflected the robustness and application potential of YOLOv8-GD method in aircraft target detection.