Improvement of multiscale feature fusion of YOLO algorithm and its application in small target detection

Aiming at the shortcomings of YOLO algorithm in small target detection, this study proposes an improved method based on multi-scale feature fusion. By designing a new feature fusion framework, optimizing feature extraction and fusion modules, and adjusting loss functions and training strategies, the algorithm's detection performance for small targets is significantly improved. The improved framework introduces an adaptive feature weighting mechanism and a bidirectional fusion path to enhance the feature expression ability of small targets; the feature extraction module optimizes the computational efficiency and detection accuracy through residual cross-layer connections and depthwise separable convolutions; the loss function adopts a dynamic weighting mechanism to increase the attention to small targets. Experimental results show that the improved algorithm outperforms the original YOLO algorithm and other mainstream methods in terms of average precision mean and small target detection indicators, while maintaining a high degree of real-time performance. These results verify the key role of multi -scale feature fusion in improving the detection effect of small targets, and provide an important reference for the further development of target detection technology.