Research On Object Detection Of Insulator And Spacer Based On UAV Vision

Power inspection is an important means to ensure the normal operation of power lines.UAV(Unmanned Aerial Vehicle)plays an important role in power inspection.When inspectors use UAVs to inspect power lines,they can use the cameras carried by the drones to collect images of the lines.This not only improves the efficiency of inspections but also guarantees the safety of inspectors.In most cases,the UAV collects data and sends it to the data center.Then,the inspection personnel will detect and compare the data to confirm the defect location of the power equipment.This off-line inspection method will increase the workload of the inspectors and reduce the efficiency of inspection.The online inspection will be done directly on the drone,then the inspection results will then be transmitted back.Online inspection will further improve the efficiency of UAV based power inspection.Current vision-based object detection algorithms are usually power consuming and therfore,require the high-performance computer,which is difficult to be deployed on UAVs.Realizing online detection on UAV with limited computing resources is a challenging problem.To solve the above problems,this article use inspection images to study the target detection of insulators and spacers in power inspections based on deep learning.The main work of this paper is as follows.1.For the power inspection tasks of insulators and spacer,the UAV is used to capture image data and built a new dataset for model training and performance testing.2.Propose a lightweight SSD network based on feature fusion.First,the lightweight network Mnas Net is used to improve the feature extraction network.Then,a feature fusion module is added to fuse the extracted multi-scale feature maps.Last,the algorithm is improved in terms of robustness by exploiting contextual information.This article tests the improved algorithm on the constructed dataset.The recognition accuracy of the algorithm is 93.8% and it takes 88 ms to detect a single image as running on NVIDIA Jetson TX2.The speed in video detection can reach 12FPS(Frame Per Second).3.A detection algorithm for insulators and spacer is proposed based on the improved siamese network.This article uses Mnas Net and Mobilenet V3 to build the siamese network to extract multi-level features.Then the features extracted by the two networks are fused to obtain a feature with stronger expressive ability.This improves the disadvantage that extracted features are usually monotonous when uses a single convolutional network.The algorithm based on the improved siamese network achieves up to 96.9% accuracy on the dataset.4.Built a UAV platform for online inspection of power inspection.This article designs the hardware structure of the UAV and configures the UAV software.The performance of the designed algorithm is tested on this UAV platform.The results show that the algorithm proposed in this paper meets the real-time requirements and can be deployed on UAV to achieve online detection.