Improved YOLOv8 for Small Aircraft Detection in SAR Images

There are many small and medium-sized object instances and strong background interference in SAR image, which is easy to cause missed detection and misdetection. To solve this problem, this paper proposes an aircraft detection algorithm based on improved YOLOv8. Firstly, the YOLOv8 network is reconstructed to add shallow feature detection heads to retain more details and enhance the algorithm's ability to capture small objects. Secondly, the C2f_DualConv module is designed to improve the information processing and feature extraction ability of the model while reducing the calculation cost and the number of parameters. Thirdly, Slim-Neck structure is introduced to improve the feature utilization efficiency of the model. Finally, the ASFF mechanism is used to improve the detection head and improve the scale invariance of the features. Experimental results on the SAR-AIRcraft-1.0 dataset show that the precision rate, recall rate and mAP50 of the proposed algorithm are respectively improved by 3.1%, 1.1% and 0.9% compared with the original YOLOv8 algorithm. Compared with the current mainstream object detection algorithms, the proposed algorithm can effectively improve the detection accuracy, and significantly improve the missed and false detection of aircrafts, and is more suitable for the task of small aircraft detection in SAR images.