Research On The Application Of Convolutional Neural Network In Infrared Image Recognition Of Power Equipment

Infrared image recognition of power equipment is a necessary prerequisite for monitoring and diagnosis of power equipment based on infrared imaging.In the actual scene,due to the imaging characteristics of infrared images,the complexity of the background environment,the diversity and difference of the power equipment itself,it is difficult to recognize power equipment in infrared image,and the traditional image recognition method based on handcrafted features is difficult to meet the requirements of power equipment recognition in the actual scene.In recent years,the deep learning technology,especially convolutional neural network,is superior to the traditional method in the performance of image recognition because of its ability to extract multi-level features from shallow to deep and perform end-to-end learning.Therefore,the application of convolution neural network in infrared image recognition of power equipment is studied in this paper.Firstly,the difficulties of power equipment recognition based on convolution neural network are analyzed.Then,in view of the difficulties,Faster R-CNN model based on Feature Pyramid Network(referred to as FPFRCNN in this paper)is selected as the research basis.Then,by analyzing the shortcomings of FP-FRCNN in infrared image recognition of power equipment,this paper improved the FPFRCNN model.Finally,the power equipment recognition algorithm based on improved FPFRCNN model achieves high accuracy and fast infrared image recognition of power equipment.The main contents of this paper are as follows:(1)Research on infrared image recognition algorithm of power equipment based on FPFRCNN model.In order to solve the problem that small equipment is difficult to recognize and the visual features of power equipment in infrared images are difficult to extract,this paper selects the FP-FRCNN model which has strong recognition ability to small objects as the research basis.By analyzing the shortcomings of the original FP-FRCNN model in power equipment recognition of infrared image,the paper improves the model by using denselyconnected structure and squeeze-excitation structure in the backbone network of the model in order to enhance the feature extraction ability of the model,and using Ro IAlign to replace the original Ro I Pooling in order to improve the target positioning accuracy of the model.The experimental results on the self-built infrared image dataset of power equipment show that the improved FP-FRCNN model can achieve 90.4% of the mean Average Precision.Compared with the original FP-FRCNN model,the improved FP-FRCNN model can increase 5.7 percentage points,and achieve a faster recognition speed of 15 fps on the GPU.(2)Design and implementation of infrared image recognition system of power equipment.In order to apply the research results of this paper directly to practice,this paper also designs an infrared image recognition system of power equipment.It consists of two parts: the main local recognition system and the remote recognition subsystem.It can accurately recognize the infrared images of power equipment stored locally or transmitted through the network,and visualize the recognition results.At the same time,it can display the recognition time,the number of targets and other related information.The system has a certain practical value.