Research On Aerial Insulator Detection Method And Its Application Based On Deep Convolutional Neural Network

In recent years,aerial UAV and computer vision technology have been widely used in power inspection,which can effectively improve the inspection efficiency of power supply bureau on transmission lines.As an important component of transmission lines,accurate inspection of aerial insulators is an important prerequisite for their defect identification and fault diagnosis.Although the traditional target detection algorithm can detect the insulator,the insulator in the aerial image is prone to miss detection and misjudgment due to the influence of occlusion,small target,different illumination,uncertain directionality,complex background and other factors,which cannot meet the current application requirements.In order to solve the above problems,this paper takes the aerial insulator as the research object,mainly uses the deep convolutional neural network to detect the insulator targets on the aerial image,so as to improve the detection accuracy of the insulator in the complex environment,and applies this method to the insulator image detection and management system.In this paper,the following work has been carried out on the issue of aerial insulator image detection:(1)In order to accurately detect and classify insulators in the inspection images of distribution overhead transmission lines,a multi-scale dense network-based insulator detection method for distribution overhead transmission lines is proposed for the application scenario of distribution overhead transmission lines.Firstly,the K-means algorithm is used to analyze the target frame of the dataset and obtain suitable anchor frames.Then,the residual module in the base network is replaced with a dense connection module to enhance the reuse and fusion of the network feature information,while a spatial pyramid pooling module is added and the multi-scale feature fusion is optimized for insulator prediction.Finally,the original loss function is updated using a focal loss function.The experimental results show that the algorithm can accurately detect the insulators on the overhead lines of the distribution network.(2)Aiming at the problem that the shape of insulator strings in the inspection images of transmission lines in the main network is uncertain in direction,and the insulator strings cannot be accurately detected in a complex background,a multi-angle candidate regional network insulator detection method is proposed.Firstly,feature information of different layers is fused to obtain more abundant features.Then,a multiangle factor is added based on the candidate regional network,and the NMS is finetuned so that it can generate candidate boxes in any direction to better fit the target.Meanwhile,the initial pooling layer is changed to the Ro IAlign layer.Finally,a 1×1convolution layer is added in front of the full connection layer of the network to reduce the parameters of its feature graph and avoid overfitting.The experimental results show that the final detection result of this algorithm can accurately detect the insulator target,and its performance is better than the current mainstream target detection algorithms.(3)An image detection and management system platform for transmission line insulators is designed and developed,and the detection algorithm is transplanted into the system to verify the effectiveness and practicability of the algorithm.This detection system platform can automatically detect and locate the insulators on the aerial images of transmission lines,and can solve the complex practical problems in the inspection of power grid.Therefore,it has high value for practical engineering applications.