Deep learning-driven machine vision system for real-time safety hazard recognition in tobacco production environments
Tobacco production improves industrial safety through computer vision and deep learning. Tobacco processing plants use end-to-end machine vision systems to detect hazardous conditions in real-time. A comprehensive study on operational risk situations has begun, constructing an annotated dataset that includes various hazardous cases occurring in different enterprises or factories under different lighting conditions. This paper provides an improved optimized backbone structure for the backpropagation convolutional neural network. Add environment-dependent and adaptive preprocessing activation functions to improve detection accuracy. Experiments have shown that it can meet strict standards under varying degrees of severe hazards, proving its effectiveness. The results indicate that compared to known benchmarks, it has higher detection accuracy, recall rate, and robustness in various high-complexity or low-visibility environments. Practical applications have shown that the system has an effective response capability to threats, capable of reducing false positives and intervening quickly. Based on the above research results, it can be predicted that improvements in machine vision-based intelligent systems will enhance industrial safety and be widely used across various industries. Develop fundamental methods based on computer vision technology to automate and optimize safety management systems in the ever-changing industrial environment.