Real Time Detection Model for Industrial Applications

With the widespread application of deep learning-based object detection technology in the industrial field, end-to-end detection models have occupied the mainstream of industrial intelligent detection due to their advantage of fast computation results. The current YOLOv8 model has been applied in various fields, and its improved optimization models are emerging endlessly. The improved models that introduce attention mechanisms often lack attention to channel information or spatial information of feature maps, and some models cannot well focus on global information, which makes it difficult to establish long-term dependence relationships. At the same time, the feature extraction and attention capabilities for targets of different scales still need further research. For this purpose, this paper proposes an improved model based on YOLOv8n. It locally introduces the Swin-Transformer structure to replace the original feature extraction module, and at the same time, it introduces the PSA structure between the Backbone and Neck structures of the model to improve the model's attention capability to information of different scales. Through comparative experiments on the PASCAL VOC2012 dataset, the analysis of the detection experimental results shows that the improved model YOLOv8n-STP network has better detection accuracy and performance.