Enhancing safety detection model for chemical laboratories in constrained computing environments
To achieve intelligent safety prevention and control in chemical laboratories using artificial intelligence and object detection technology,we propose an enhanced object detection model.This model,based on the improved YOLOv8 architecture,aims to detect irregular behavior of laboratory personnel and abnormal laboratory conditions.The conventional object detection models are known for their complexity and high computational demands,which pose challenges for typical chemical laboratories.Therefore,we have opted for the YOLOv8n model,a smaller-scale solution,and implemented lightweight enhancements.Specifically,we utilized Group Shuffle Convolution(GSConv)to reconstruct the C2f structure of YOLOv8 and the downsampling module in the feature fusion section.These modifications aim to decrease the model's parameter size and computational complexity,ensuring efficient performance in constrained computing environments.This enhances the computational efficiency and inference speed of the model,thus better meeting the real-time requirements for safety detection in chemical laboratory scenarios.Additionally,we designed a lightweight detection head based on group convolution to replace the original detection head of the YOLOv8 network,further reducing computational complexity while maintaining the model's detection performance.Due to the scarcity of safety detection datasets tailored specifically for chemical laboratory scenarios,we collected numerous images depicting abnormal situations and unsafe behaviors of laboratory personnel through online sourcing and on-site photography.These images underwent detailed annotation using labeling software,particularly LabelImg.Subsequently,we curated a specialized safety detection image dataset designed specifically for chemical laboratory settings,which was then utilized for training our model.We conducted ablation experiments on the enhanced modules within the same experimental environment.The results demonstrated that our improved model achieved a 48％reduction in parameter size and a 54％decrease in computational complexity,all while maintaining high detection accuracy.With a size of approximately 3 MB,our enhanced model achieved a precision of around 80％,striking an optimal balance between performance and lightweight design,making it highly suitable for deployment in resource-constrained environments.Our research contributes new insights into intelligent safety prevention and control in chemical laboratories and provides a reference dataset.Furthermore,our improved model demonstrates broad application prospects in other scenarios requiring lightweight and efficient security monitoring.
万方数据
维普
