Research On Semantic Segmentation Of Power Line Based On Image

Semantic segmentation of power line is a very pivotal task in the field of digital image processing,which can be used as a preliminary task for many tasks,such as ice detection of electric transmission line in the wild,detection of trees or engineering vehicles' intrusion,detection and mapping of electrical components on wires,disaster monitoring and etc.In addition,semantic segmentation of power line can also directly provide navigation for line-patrolling robots or unmanned aerial vehicle applied in power systems.Detection of power line,i.e.semantic segmentation of power line,is an important and challenging research in the crossing fields of digital image processing,computer vision and power system engineering.Firstly,the algorithm and technology of edge feature and line feature extraction in the field of digital image processing are studied.In this paper,the classical edge feature extraction methods and line detection algorithms are introduced in detail.The advantages and disadvantages of these methods are analyzed.The application effect of these methods in transmission line detection is explored by experiments.The Haar-like feature is extended and improved in the paper.The paper studies the application of neural network of semantic segmentation based on deep learning technology in power line detection.This paper introduces in detail the design ideas and specific methods of several current mainstream convolutional neural networks of semantic segmentation such as FCN,SegNet,DeepLab series,and analyses the main modules in these networks,such as ASPP module,full-connection conditional random field,etc.Convolutional neural network's advantages and disadvantages in the application of semantic segmentation of electric transmission line are analyzed by experiments.In this paper,a more complex decoder structure is designed on the basis of DeepLab v3+ neural network,which makes DeepLab v3+ neural network more suitable for semantic segmentation of power line.The paper also studies the combination of edge feature extraction algorithm based on digital image processing technology and convolutional neural network of semantic segmentation based on deep learning technology.This paper explores two different ways of combining Haar-like features with DeepLab v3+ network.In this paper,the two methods are applied to power line detection,and the experiments prove that both methods are helpful to improve the accuracy of wire segmentation,which is superior to the traditional digital image processing method and the neural network method.At present,the application of traditional digital image processing technology and computer vision technology based on deep learning neural network are relatively independent.I hope that the exploratory research in this paper can provide some references for later researches on how to better integrate traditional digital image processing technology and deep learning algorithm,especially in engineering applications.