In this paper,we aim to improve the accuracy of the flying bird detection task,and propose an improved YOLOv8 algorithm based on deep learning to address the shortcomings of traditional algorithms in the detection of flying birds with complex backgrounds and multi-variable attitudes.By fusing the GD mechanism in GOLD-YOLO and the Scale Sequence Feature Fusion(SSFF)module in ASF-YOLO,the network structure is optimized,the feature extraction and fusion methods are improved,the multi-scale feature fusion capability is enhanced,and the feature representation of small-size targets is improved.In addition,a fly-ing bird dataset containing close-up photos of different birds,flying birds and flock images is collected to support the training and testing of the algorithm.The experimental results show that the improved YOLOv8 algorithm improves the performance in the flying bird detection task,especially when dealing with small targets and complex backgrounds.
万方数据