Intelligent identification method of mine fire video images based on YOLOv5

In order to solve the problems of video image distortion caused by uneven light distribution and low accuracy of fire identification in coal mines, an intelligent identification method of mine fire video images is proposed. The method uses YOLOv5 as the identification model and uses K-means algorithm to improve the traditional dark channel image defogging algorithm to defog the collected flame images and improve the identification accuracy of mine fire video images. In order to reduce the impact of static background on fire identification, the fusion algorithm of frame difference method and Gaussian mixture model is used to extract the characteristics of the dynamically evolved flame images, and the morphological processing algorithm is used to eliminate the gaps in the images so as to obtain more complete flame target images. The fire video image data set is annotated and input to the YOLOv5 algorithm model for training and testing. The results show that the average accuracy of the intelligent identification method of mine fire video images based on YOLOv5 is 92% with a loss function of 0.6, which is 9.6%, 13.5% and 4.9% higher than that of the traditional algorithms, Alexnet, VGG16 and Inceptionv3 respectively, indicating that this method has fast detection speed and high accuracy, and can improve the accuracy of mine fire identification effectively.