The filter bag of the bag filter in the dry dust removal device will be damaged after long-term use, resulting in serious problems such as increased energy consumption, reduced dust removal efficiency, and environmental pollution. Therefore, a dust and smoke detection webcam is built at the dust removal chamber, and the leakage of smoke at the hole is detected based on the YOLOv3 algorithm. The test and detection steps are as follows: 1) According to the size of the real dust removal chamber, the smoke leakage laboratory platform at the hole of the bag filter is designed, and the image data of smoke leakage at different holes are collected; 2) A software is used to label these image data; 3) The Darknet deep learning framework is built, the YOLOv3 algorithm is used to train the image data, and the broken bag condition is identified according to the calculation results of the model. The results show that the recognition accuracy of the model can reach more than 91% for batch images and 95.56% for continuous videos. The system in this paper can reduce the manual detection in the operation of the dust collector, reduce the cost of human resources and the labor intensity of workers, avoid potential safety hazards and losses in production, provide technical guidance for the production safety and regular maintenance of the factory, and also provide effective solutions for the intelligent production of the whole plant.