MOPE-YOLO: Improved Laboratory Safety Detection Algorithm Based on YOLOv8

The role of personal protective equipment is crucial in laboratory safety. However, the current mainstream detection algorithms are clearly inadequate in dealing with the testing of protective equipment, and cannot solve problems such as image occlusion and difficulty in edge deployment. In view of this, this article innovatively proposes a protective equipment detection algorithm based on YOLOv8 framework optimization - MOPE-YOLO. Firstly, in order to balance the conflict between the increase in model parameters and low floating point operations (FLOPs), the GM-DynamicConv module was carefully designed to replace the original C2f convolution. Secondly, considering the limitations of model edge deployment, we have introduced a newly designed SPPF_H module to replace the original SPPF module. This improvement maintains the dual advantages of model computational efficiency and detection performance. Finally, considering the loss caused by image occlusion areas, an innovative SEMA_H convolutional attention module suitable for the head is proposed. This module perfectly enhances the model's ability to detect occluded objects through fine attention regulation of scale, space, and channel dimensions. Through experimental verification, the MOPE-YOLO algorithm has achieved significant surpassing in multiple key evaluation indicators, including an accuracy improvement of 5.9%, a recall increase of 0.6% and [email protected] increased by 1.3% and the computational complexity decreased by 2.0 GFLOPs.