Research on Application of Target Detection Network Based on SSD in Underground Coal Mine

Aiming at the problems of poor underground coal mine environment, poor illumination, mixed background, blurred pedestrians, multi-scale pedestrians, etc., and traditional target detection usually extracts features manually, which requires a large workload, cumbersome detection steps, low detection accuracy, and serious missed detection. Based on the theory of deep learning, this paper proposes a method for identifying people underground in coal mines based on Single Shot MultiBox Detector (SSD). According to the self-made downhole worker data set based on the VOC data set, the labelme labeling tool is used to manually label the downhole personnel as "person", and the SSD-VGG16 network model is trained. Experiments show that the model can effectively detect people underground in coal mines. When the IOU threshold is 0.5, the SSD model has an average accuracy of 97.13% and a recall rate of 87.53%. The model can better detect underground personnel and has a certain The universal applicability of the mine is of reference significance for the intelligent monitoring system of coal mines.