Research On Insulator Detection Based On Deep Learning

The high-voltage transmission network is one of the most important infrastructures in our country,and insulators are an important part of the high-voltage transmission network.The health of the insulator is an important factor affecting the normal operation of the transmission network.Due to prolonged exposure to the natural environment,insulators are prone to deformation,cracks and even breakage,which will adversely affect the stable operation of the transmission network.Therefore,regular inspections of the transmission network are necessary.With the popularity of drones,drone intelligent inspections are gradually replacing manual inspections due to their low cost and high safety.How to automatically identify and classify insulator pictures taken by drones has become the current situation.Research hotspots.This paper is based on the deep learning model to study the detection of insulators.First of all,this paper improves the YOLOv4 model with the goal of improving the positioning accuracy.Integrate the compression-stimulation attention network in the feature extraction network of the YOLOv4 model.By compressing and stimulating the two states,the model automatically learns the importance of features between channels and obtains the weight of each channel,thereby improving the detection effect.Secondly,the paper proposes an improved YOLOv4 model based on the IoU-Aware mechanism.Since YOLOv4 is a one-stage target detection algorithm,classification and positioning are independent of each other.Due to the lack of cross-comparison information during inference and prediction,only the classification scores are used for sorting,which may cause mismatches between classification and positioning,resulting in some anchor points with higher cross-comparison scores but lower classification scores being mistakenly Filtered.In order to solve the above problems,the paper introduces the IoU-aware mechanism in the original YOLOv4 network,that is,the intersection ratio score is also used as a consideration factor for sorting during non-maximum suppression during reasoning,so as to improve the positioning accuracy of the network.Finally,in view of the problem that the type and form of insulators in the existing public data sets are too single,the paper expands the insulator data set.In the research process,different types of insulators were collected through the Internet and manually labeled,and the public data set was expanded by flipping and adjusting the contrast.The YOLOv4 insulator detection algorithm model based on compression-inspiring attention and IoU-aware proposed in this paper is trained on the expanded data set.The experimental results show that the YOLOv4 insulator detection algorithm model based on compression-inspiring attention and IoU-aware proposed in this paper improves the detection effect of the original YOLOv4 algorithm model.