| The land area of China,which is the third largest in the world,determines the characteristics of long-distance transmission distribution of transmission lines in China,and ensuring the safety of transmission lines is a necessary condition to protect China’s national economic life.The occurrence of mountain fire will occur line trip,leading to irreversible large-scale accidents,so timely detection of mountain fire is particularly important.In the early stage of mountain fire,its obvious feature is smoke,and the study of fire smoke detection technology is a necessary means to protect the safety of transmission lines.This paper studies the transmission line fire smoke detection technology based on image and video.Taking the mountain fire smoke of transmission lines as the research object,firstly,the image-based method is adopted to detect and analyze the mountain fire images collected by monitoring,and the characteristics of the color and texture characteristics of the smoke are introduced and analyzed,as well as the principle and application of neural network related methods.Then a lightweight convolutional neural network framework for smoke detection of transmission lines is constructed to realize smoke detection and classification.Secondly,it is found in the actual detection that foggy days have a great impact on the accuracy of mountain fire smoke detection,so an image haze enhancement algorithm based on local and global image is proposed.Local image processing was carried out by image equalization,and global image enhancement was carried out by improved single-scale Retinex method,which ensured the overall defogging effect and clarified the boundary of smoke.In order to make up for the shortage of image samples in foggy days,an image style transfer method combining texture synthesis and color transfer is proposed.Taking the fog image as reference,the original sample set of mountain fire smoke image was expanded into fogging data set,and finally combined with the neural network of mountain fire smoke detection to improve the accuracy of the algorithm.Due to the fact that fixed cameras are mostly used for shooting and monitoring in actual detection,the video-based method provides more coherent information,which can be used to detect the early small and thin mountain fire smoke,this paper further studies the video based method on the image.Firstly,the smoke grid features with color as the prominent feature are studied and designed.According to the structural coherence characteristics of the transmission line surveillance video images,the grid features consistent with the area of the smoke itself are obtained.Secondly,the method of obtaining early mountain fire smoke candidate frame based on anomaly detection is proposed.By designing the overall flow frame of anomaly detection algorithm,anomaly detection is carried out on the information of the front and back frames of the smoke video,and the areas that meet the judgment conditions are marked as smoke candidate frame.Finally,the improved local binary mode operator is studied to enhance the image texture information,and it is applied to the candidate box,combined with support vector machine to detect the improved texture features.The video-based method has a good detection effect on early small and thin smoke.The two methods of detecting mountain fire smoke on transmission lines based on image and video can achieve good detection effect,which is of great significance to the safe production of transmission lines. |
