Research On Power Equipment Recognition And Thermal Anomaly Detection Method Based On Computer Vision

The normal and stable operation of power equipment is the basis for the long-term stable work of the power system and is closely related to the daily production and life of the people.Therefore,ensuring the normal operation of power equipment is of great significance.However,traditional manual inspection methods have disadvantages such as low efficiency and failure to find hidden dangers in time.The development of technologies such as computer vision and deep learning is becoming more and more mature.Applying them to power equipment identification and thermal state monitoring can effectively improve the efficiency of inspections,realize intelligent,automated,and informatized inspections,and have broad application prospects.Aiming at the problems of power equipment identification and thermal state detection in the intelligent inspection of power equipment,this paper studies image processing technologies such as image denoising,image enhancement,feature extraction and matching and deep learning theory in the field of computer vision,and proposes a feature-based matching The power equipment identification method and the thermal anomaly detection method based on the deep learning target detection model have laid the foundation for the realization of intelligent power equipment inspection.The main research contents of this thesis are:(1)The feature matching method is used to realize the identification of power equipment in the video monitoring collected images.Aiming at the shortcomings of the accelerated robust feature(SURF)algorithm,distance constraint conditions are used to optimize,and feature matching is performed according to the principle of similarity measurement,and the combination is immediately consistent The performance algorithm filters out the wrong matching pairs,and locates and recognizes the template power equipment in the image to be recognized.After experimental testing,the accuracy rate reached 93.4%,which verified the effectiveness of the method.(2)According to the type of data required by the target detection model in deep learning,a power equipment infrared image data set is constructed,and data augmentation technology is used to expand the data set to provide a basis for subsequent research on thermal anomaly detection methods.(3)Deeply studied the convolutional neural network and target detection model in deep learning theory,verified the feasibility of the target detection model applied to the thermal anomaly detection of power equipment through experiments,and compared the test results,weighing the accuracy and detection speed.Based on the YOLOv3 target detection model,the network feature extraction structure is optimized through the dense connection method,and the soft NMS algorithm is used to optimize the border regression.A higher recall rate is achieved,and the accuracy rate of the original model is increased by 3.4%.It is a deep learning in the power system The application of thermal anomaly detection provides certain reference value and significance.