Search

Detection of line component defects in UAV transmission line inspection is always a difficult problem. In order to solve the problem of identifying the defects of insulators and anti-vibration hammers in transmission lines, a target detection technique based on deep learning is proposed to diagnose the typical defects of transmission lines. SSD algorithm based on candidate regions is selected to locate and identify the defects. Firstly, the influence of different feature extraction networks and network parameters on the accuracy and speed of target detection is studied. The network is improved by adjusting network parameters and model optimization method, and the network parameters that are most conducive to transmission line defect detection are selected.Then, the influence of different data enhancement methods on the accuracy of defect detection is studied. The training samples are expanded by multi-scale training and horizontal mirror method to further improve the accuracy of target detection.Finally, the recognition and classification experiments are carried out using the actual images collected by UAV. The experimental results show that the target detection method based on deep learning can accurately locate the location of defects from the transmission line image, and can be applied to the fault diagnosis task in the transmission line scene.

With the increase of the transmission line mileage, the transmission line inspection is facing severe challenges. As an efficient inspection way, the UAV (Unnamed Aerial Vechicle) is used for the transmission line inspection. However, at the present stage, to complete the observation and data collection of transmission line equipment, the PTZ (Pan Tilt Zoom) camera in the UAV is completely controlled dependent on the operator during the inspection. These needs higher requirements for the operator and lead to high labor intensity. And the image scale of the collection device is not uniform, which improves the difficulty of subsequent intelligent diagnosis of defects. In order to solve this problem, this paper designs an intelligent acquisition system for UAVs based on edge computing using in the transmission line inspection. The system includes the front-end edge computing detection module based on SSD (Single Shot Multibox Detector) algorithm and a PTZ camera control module. The system workflow is as follows: firstly, the UAV acquires the video stream data of the surrounding environment of the transmission line through the fixed-focus camera in the PTZ. The front-end edge computing detection module receives the video stream through the USB interface. And the front-end edge computing module uses the SSD algorithm to identify the video stream gotten from the PTZ camera to recognize and locate the transmission line equipment. Secondly, if the transmission line equipment is identified in the field of view, such as an insulator, the anti-vibration hammer, the equalizing ring, etc., the front-end edge computing module calculates the target image ratio of the device and from the center position based on the identified pixel position information of the transmission line equipment, and converts to the pan-tilt rotation amount and camera pulling factor. The pan-tilt rotation amount and camera pulling factor is transmitted to the PTZ camera control module through the serial port. The PTZ camera control module controls the camera's movement until the transmission line equipment is in the center of the field of view. Finally, the wide-angle camera in dual-view PTZ camera is controlled to zoom. And the image after the zoom is photographed and saved to finish a flow of the data of transmission line equipment collects. After testing, the SSD algorithm for transmission line equipment can achieve an overall recognition accuracy of 73%. The insulator identification accuracy is 75.7%.The anti-vibration hammer identification accuracy is 72.6%.And the equalization ring recognition accuracy is 70.3%.The acquisition video is 1080P, the edge computing module processing speed is 3fp/s, and the pan/tilt control module controls the pan/tilt camera to complete a shooting time of 3s. On the one hand, the system realizes the autonomy, process and standardized collection of the image information during the transmission line equipment while improving the validity of the image data information and reduces the complexity of the subsequent defect diagnosis. On the other hand, it also reduces the manual participation degree of the inspection operation so that the efficiency can be increased by 5~7 times compared with the manual control of the PTZ camera for equipment acquisition.

In this paper, a fire detection method based on color features, wavelet analysis, and convolution neural network is proposed. Firstly, the candidate region of flame is extracted by color segmentation method, and then the candidate region of smoke is generated by the background fuzzy model based on wavelet analysis. Then the candidate region is filtered by the trained CNN model, and the position of flame and smoke in a picture is located. Finally, a large number of fire pictures in different scenes are used to test the algorithm. The results show that this method can detect the location of flame and smoke accurately and quickly from images or videos, and can be applied to fire detection tasks in substation scenarios.

This paper presents an intelligent defect diagnosis algorithm based on fully convolutional neural network for typical transmission line components. Based on the region-based fully convolutional neural network algorithm and combined with deformable convolution, feature context fusion, clustering, the feature expression ability of neural network is improved. Through the improvement, the algorithm adapts to the object deformation and scale difference. In addition, the training effect is improved by improving the sample labeling strategy. In this paper, the defect diagnosis of four kinds of transmission line components(insulator, vibration damper, grading ring, wire clamp) is realized.