Research And Implementation On Infrared Image Segmentation Technology Of Electrical Equipment Based On Mask R-CNN

Infrared thermal imaging technology has the characteristics of non-contact temperature measurement,and has been widely used in thermal fault detection of electrical equipment.At present,the analysis of infrared images mostly depends on human.With the continuous expansion of the scale of power construction and the rapid increase in the number of power equipment,limited manual methods cannot meet the increasing detection needs,and it is urgent to use automated methods instead.In recent years,artificial intelligence-based deep learning algorithms have developed rapidly and have been widely used in many fields in the electrical industry.From the perspective of infrared image recognition,this paper studies the electrical equipment detection technology based on deep learning.This paper contains the following specific work:Aiming at the current lack of research on infrared heat map processing in power equipment monitoring,combined with the development trend of intelligent monitoring of electrical equipment,a multi-target detection and segmentation task of infrared images of equipment is proposed;By comparing the respective characteristics of traditional and deep learning image processing technologies,a deep learning method suitable for multi-object detection and complex background segmentation tasks is selected,and a Mask R-CNN model is selected for research;Aiming at the shortcomings of the Mask R-CNN in image segmentation,an improvement scheme is proposed.On the basis of the original model,strengthen the fusion of low-level feature information,and use the features of each layer of the feature pyramid as the feature values of RoI,and finally merge the feature information of the fully convolutional network and the fully connected layer in the Mask branch;Comparative experiments are performed on the original and improved Mask R-CNN models.First,the experimental data set is preprocessed and labeled,and divided into a training set and a verification set according to a ratio of 8:2,and then the training set is input to the model training and verified using the verification set.Finally,the model performance is compared according to the official COCO evaluation metrics.The experimental results show that the improved Mask R-CNN improves the segmentation ability and performs best in the infrared image processing task of electrical equipment in this paper.