MF-YOLO: Mask Wearing Detection Algorithm for Dense Environments

To address the challenges of false positives and missed detections in face mask detection within dense environments, we propose the MF-YOLO face mask detection model. First, a feature map convolution approach is introduced to transform the feature map information into the initial weights of convolutional kernels, accelerating the convergence of model training. Additionally, a residual self-attention module is designed to capture fine edge position features by combining global feature extraction with sliding window-based information interaction, effectively reducing missed detections, especially for small objects. Furthermore, we introduce a mixed attention mechanism that integrates both channel and spatial attention, enabling the model to focus more effectively on key features and better target the objects of interest. Finally, an enhanced feature-fusion strategy is employed to mitigate the loss of feature information during layer-by-layer transmission, improving the model’s ability to fuse information at the same scale. Experimental results in the public MASK dataset and the self-built SFMD dataset demonstrate that the proposed algorithm significantly accelerates convergence while improving accuracy by 2.1% and 1.3%, respectively, compared to the original network. Our code and datasets are publicly released on GitHub <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/wenpengshishuaige/MF-YOLO</uri>.