Research On Power Line Inspection Based On UAV Power Inspection

The power line is an important part of the electric transmission line.Due to long-term exposure in the natural environment,it is prone to face challenges such as vegetation invasion and foreign body suspension,which cause great hidden dangers that affect the safe operation of the power line.With the development of science and technology,UAV inspection is becoming the main method of power inspection.Power line inspection plays an important role in auxiliary positioning and path planning of the UAV so that benefit the fault detection during a UAV based transmission line inspection.However,current power line detection algorithms have problems such as slow detection speed,inability to detect on multiple scales,and low detection accuracy in complex scenes.In addition,the lack of public power line datasets makes it difficult to judge the performance of the algorithm.In this paper,around these issues,a UAV inspection platform is built,and the power line data is collected to calibrate the two power line data sets of multi-scale and multi-scene.It further studied the features of the power line and proposed a parallel line detection-based power line extraction method using edge detection,which can effectively overcome the problem of power line detection on multiple scales.A power line detection method based on deep learning semantic segmentation network is proposed,which has good detection accuracy and fast detection ability in complex scenes.The specific research content of this article is as follows.First of all,in terms of UAV power inspection,a quadrotor UAV inspection platform was built based on the open source hardware Pixhawk and the open source software Ardupilot.A large amount of power line data was collected through the built UAV inspection platform,and two power line data sets were calibrated,namely,a multi-scale power line data set and a multi-scenario power line dataset.Aiming at the problem of power line texture change,a power line extraction algorithm based on parallel line detection and straight line extension is proposed.The algorithm first extracts the image edge through the Canny algorithm and then constructs a soft voting function for power line extraction based on the length,width,and direction of the power line characteristics.Finally,for the candidate power lines extracted by voting,the precise power line area is obtained after optimization processing such as parallel line extension and connection.The algorithm was verified on a calibrated multi-scale power line data set,and the results showed that the algorithm can effectively overcome the power line detection problem caused by texture changes,and it still has a high accuracy rate when the UAV is far away from the power line.Aiming at the power line detection problem in complex scenes,a power line extraction method based on improved LinkNet is proposed.The algorithm extracts power line features through the input layer of the Res Net network and adds a hole convolution block to constrain the power line structure features.The class-balanced cross-entropy loss function is used to weight the loss according to the number of corresponding pixels to overcome the imbalance in the proportion of positive samples and negative samples in the power line sample image.The effects of different upsampling methods on the power line extraction results are tested to select a suitable network structure.The performance of the algorithm is tested on a multi-scenario power line data set.Experiments prove that the power line extraction algorithm based on deep learning can overcome the influence of complex scenes and accurately extract power lines.Finally,the power line detection algorithm based on deep learning was tested on a mobile platform,which verified the feasibility of the algorithm for deployment on UAV equipment.