Research On Power Lines Detection Algorithm Based On Deep Learning

The power lines that exist everywhere in life can cause some troubles to us while delivering electricity.In low-altitude flights,accidents are caused by neglecting the existence of power lines.The long-term using of the transmission line causes many problems,therefore,frequent maintenance is required to ensure a safe and stable operation of the transmission line.Transmission line detecting technology plays an important role in low-altitude flight safety and transmission line maintenance.The traditional detecting method apply simple prior knowledge and manual features to detect power lines.There are many limitations in accuracy and application range.The rise of deep learning,especially in the field of image analysis and video analysis,is widely used.In dealing with the transmission line detection by directly applying the existing deep learning method,however,there are no satisfactory effects.Based on the existing deep learning algoritm,this paper designs and implements two transmission line detection methods for the application of line anomaly detection in low-altitude flight safety and transmission line maintenance.In terms of low-altitude flight safety,transmission line detection mainly solves the problems of "line presence" and "line distribution".In this paper,the semantic segmentation method is used to classify the input image at the pixel level.Finally,the transmission line detection result can be obtained while providing the specific position of the transmission line in the image and the relatively accurate contour information.In deep learning,algorithms are driven by a large amount of training data.As there is no open source transmission line detection data available,this paper constructs a semantic segmentation data set for power lines.At the same time,five sets of comparative experiments and analysis were carried out for exploring the appropriate receptive field size of the semantic segmentation network in the process of detecting lines.In terms of line maintenance,the transmission line detection mainly solves the problems of "line existence" and "abnormal existence",but it does not pay attention to how the line is distributed.In this paper,the input image is detected by the classic Region Proposal Network.Detection boxes provide local positioning information of the line.Finally,Detection boxes are classified to determine the abnormal type of the transmission line.Using the created semantic segmentation dataset,the dataset can be converted into a data set that are used for target detection.However,only the positioning information of the transmission line can be obtained,and there is no abnormal type tag information of the line,as a result,this paper only made an experiment on the problem of "line existence".In order to improve the detection effect of power lines,this paper achieves an improvement which is based on the RRPN network,obtaining a network suitable for abnormal detection of power lines,and making an experiment and analysis on the algorithms.This paper establishes a data set for open source network-based transmission line detection that is available for open source,and provides a research basis for future research work.This paper analyzes the main problems of transmission line detection in two different application scenarios,and responds to different detection scenarios with different detection methods.Among them,the model of transmission line detection based on semantic segmentation has a pixel accuracy rate of 0.72.Based on the RRPN transmission line detection model,the Intersection-over-Union of the detection boxes is 0.76.Both detection methods show high detection accuracy in different scenarios,showing the great potential of deep network in solving transmission line detection problems,and providing a new research direction for transmission line detection problems.