Research On Detection Algorithm Of Power Channel Violation Building Based On Deep Learning

Due to the high-voltage transmission of electrical energy and the construction of intelligent power networks,ensuring the safety of power passages has become one of the first tasks in the power sector.The phenomenon of violation buildings occupying power passages has been banned repeatedly,which has caused major hidden dangers to safe power transmission and power network safety.Most of the current mainstream deep learning building detection algorithms are aimed at the detection of violation buildings in urban remote sensing images,and the performance of video sequence images under dynamic background is not good.This article is based on the actual needs of violation building detection in power passages,and studies and applies deep learning techniques represented by convolutional neural networks for violation buildings occupying power passages to achieve accurate location of buildings in sequence images and design violation building discriminators.Achieve the goal of detecting violation buildings in power passages.The main research contents of this article:(1)The number of violation building objects is uncertain.The building detection method of classification and detection is used to pre-process the collected power channel videos.Annotate 4 types of classified datasets including buildings,power towers,vehicles,and trees,and expand the data.The original Alexnet network structure is optimized to obtain the Imp-Alexnet network,which improves network classification performance,achieves building area extraction and building classification goals,and obtains building area location information.(2)The sizes of violation building objects are different.Adopting the building detection method of detection before classification,the samples containing violation buildings are expanded by using GAN to build a power channel building detection data set,and the network performance of Faster R-CNN is optimized by adjusting network parameters and introducing a focal loss loss function.The detection of most buildings in the power channel video results in more accurate building location information,reducing missed detections and false detections.And the video input network collected one year apart on the same line channel is used to verify the effectiveness of the algorithm,and it is found that the network can accurately detect all buildings in the power line channel.(3)Design of violation building discriminator.Aiming at the needs of violation building determination,based on the building location information detected by the neural network,a distance-based violation building discriminator is designed and combined with the ORB algorithm to improve it to achieve building matching in sequence images and further improve the power channel Speed ??and accuracy of violation building detection.