Facial Detection Algorithm based on Improved YOLOv5s

To address the challenges of low detection accuracy and the ease of misdetection and omissions due to small face targets or masked faces in complex environments, this paper proposes a face detection algorithm based on an improved YOLOv5s. Initially, the EIoU bounding box loss function is introduced to enhance the regression speed and localization accuracy of the target bounding box. Next, the CBAM attention mechanism module is incorporated to augment the YOLOv5s network’s capability in extracting image features without substantially increasing the model’s parameters. Finally, the post-processing NMS algorithm is enhanced to optimize the redundant bounding box rejection method and filter out high-quality detection results. Experiments on face detection are carried out using the YOLO_Mask dataset, resulting in an enhanced Precision by 2.2%, mAP@0.5 by 2.6%, and mAP@0.5:0.95 by 4.2%. The experiments demonstrate that the improved face detection algorithm notably enhances the detection of small and occluded targets, validating the effectiveness of this improved method in addressing the issues of low detection accuracy for small and occluded targets, as well as misdetection and missed detections.