Research On Recognition Method Of Electrical Components Based On Deep Convolutional Neural Networks

In recent years,with the continuous expansion of power grid,the length of transmission lines is also increasing.In order to ensure that the electrical components are in normal state during the operation of the power grid,it is necessary to focus on the inspection of electrical components such as insulators and shockproof hammers during the daily inspection of the power grid.The traditional manual inspection method can no longer handle large-scale inspection data,and the inspection efficiency is low under the complex environmental background.UAV equipped with visible light camera can achieve the inspection of power components efficiently and safely,but it still needs to solve the problems of massive data processing and how to fully extract the features of electrical components under the complex environmental background.According to the characteristics of electrical components in high voltage transmission lines,this paper makes an in-depth research from the following aspects.(1)Based on the research of the traditional and deep learning-based image compression algorithms,this paper puts forward a hierarchical compression algorithm combined with visual attention mechanism to solve the problem that it is difficult for UAV to deal with massive inspection data in a short time.By using the Block CNN combined with the GMR to achieve the region segmentation and efficient compression of the regions of interest and non-interest regions in the inspection images,which filters out some redundant information in the inspection image and reduces the transmission time of a single image.Compared with the traditional image compression algorithms,the compression framework proposed in this paper can be effectively applied to the compression and transmission of inspection images in UAV power inspection system.(2)This paper analyzes the traditional and deep learning-based image super-resolution reconstruction algorithms in depth.In view of the problem that some inspection images show blurred quality due to UAV fuselage shaking and problems of imaging exposures,which will seriously affect the subsequent detection process of electrical components.The SRCNN network is used to achieve the super-resolution reconstruction for blurred images.It means the low-resolution inspection image is reconstructed into the high-resolution inspection image.The reconstructed clear images are combined with other training images to realize the quality improvement and expansion of the datasets,and the performance of the algorithm is compared with that of other Super Resolution algorithms.(3)Aiming at the problem that traditional algorithms cannot effectively extract the features of electrical components under the interference of complex environmental background,this paper proposes an object detection method based on YOLO V3.By training the dataset,the features of electrical components are fully extracted,and the corresponding training parameters and training model are obtained.The detection speed and accuracy of electrical components in UAV inspection images are improved.Moreover,the performance of the algorithm is compared with that of other popular object detection methods.(4)For the specific experimental details of SRCNN and YOLO V3 networks,this paper conducts further analysis on the production of dataset,the selection of image classification algorithm,data optimization algorithm and determination of network parameters.The experimental results verify that the proposed method can effectively locate the electrical components and identify their physical status.The algorithm has high recognition accuracy,good robustness and strong real-time performance,and has more advantages in the detection performance of electrical components.Therefore,the detection model can be effectively applied to the UAV intelligent power inspection system.