Image Monitoring Technology for Industrial Emissions Based on Improved YOLOv8

In industrial emissions monitoring, the complexity of environmental conditions and the close association between emissions characteristics and environmental factors make the extraction of key emissions features a prominent challenge. To address this challenge, this study introduces the BiFormer attention mechanism into the backbone network of the YOLOv8 model, significantly enhancing the model’s ability to extract various industrial emissions features, such as gases and smoke, and improving the model’s adaptability and robustness in complex industrial environments. Considering the requirements for real-time monitoring of industrial emissions and the limitations of device resources, this paper integrates partial convolutional modules (PConv) to replace the conventional 3x3 convolutional modules in the bottleneck of the feature extraction network, reducing memory access frequency and computational redundancy, and further optimizing the model’s operational speed. Experimental outcomes indicate that the enhanced YOLOv8 model boosted the mean average precision by 3.2 percentage points on the publicly available smoke dataset and decreased computational load by 15%, thereby supporting efficient operation on resource-limited devices and fulfilling the requirements of practical industrial applications.