Research On Electric Equipment Image Recognition And Its Application Based On Deep Learning

In order to improve the intelligent level of the power system,besides the four traditional functions of "remote testing","remote informing","remote controlling" and "remote regulating",nowadays in China,many substations have added the "remote viewing" function,which can remotely evaluate the operation status of the electric equipments through the image and video monitoring systems.The video and image information is very useful.However,many of the monitoring systems only can provide video and image information,and they cannot intelligently analyze or recognize the electric equipment from the multimedia stream.After receiving the unstructured mass electric equipment images,the operators in the control center have to analyze and recognize the electric equipment by themselves.This significantly increases the burden of the operators.In order to solve this problem,this paper introduces the digital image processing and deep learning technology,which can analyze and recognize the electric equipment from the multimedia stream automatically.The research contents include image pre-processing,object detection,image segmentation,features extraction and seletion,object classification and recognition.Firstly,this paper points out that it is necessary to apply the digital image processing technology to analyze and recognize the electric equipment.After overview the current research status of the image processing technique in power system,the limitations of the existed methods are outlined.Then,deep learning technology and its recent research progress are introduced,which provide the fundamental basis for further research.Then,in order to solve the problem of color image enhancement for edge information protection,a new nonlinear adaptive enhancement(NAE)algorithm is presented to resolve the problem in parallel procedure for low or high intensity and poor contrast(LIPC or HIPC)electric equipment images.The NAE algorithm consists of three steps.First,a RGB color image is converted into an intensity image.Then,an adaptive intensity adjustment with local contrast enhancement is performed parallelly,by using a single scale shift-variant Gaussian bilateral filter and the second order Taylor series expansion approximation technology.Finally,the colors are restored.A significant advantage of NAE is that the edge information of enhancedimages can be enhanced or preserved for LIPC or HIPC images.The experimental results show that the visual effects of NAE are same or better than other enhancement algorithms,and the enhanced images have more edges information changes by NAE.Next,borrowed ideas from human visual perception system,motivated by the visual attention mechanism of human beings,this paper builds a convolutional neural networks based visual saliency model,which considers both the regional features as well as the border features of the electric equipment,to detect the electric equipment automatically from the multimedia data.At first,the superpixel segmentation method is used to segment the image into several superpixles.Then,a bilateral filtering method is used to get the regional information of the electric equipment.After that,two independent CNNs are used to learn both the regional information and the edge information of the electric equipment.Lastly,the confidence coefficients are feed into the conditional random filed to calculate the saliency maps.Experimental results show that the proposed method can highlight the region of the electric equipment,and it also can suppress the background of the images correctly.Furthermore,it can detect the electric equipment and keep the edge information of the equipment efficiently.After that,in order to recognize the type of the electric equipment,this paper proposes a double channel convolutional neural network(DCNN)model to extract features for the electric equipmens.Different from the Alexnet model,which extracts image features through one channel,the DCNN model extracts features through two independent channels.The features extracted by DCNN mdoel can reflect the essential characteristics of the electric equipment,and they are much more abstract than the features extracted by the Alexnet model.Then,based on the deep learning features,random forest(RF)is applied to classify the electric equipment into different categories.Experimental results show that the recognition accuracy of the proposed method is much higher than that of the traditional random forest and the softmax-based deep convolutional neural network.Furthermore,it can effectively eliminate the effects,which are produced by the complex background.The proposed method can meet the actual command of the electric power department,and it provides a new solution for intelligent analysis and recognition of unstructured mass electric equipment images.Finnaly,considering that the traditional power operator training systems have many shortcomings,this paper proposes to apply the electric equipment recognition method into the augmented reality(AR)intelligent assistant system,and it develops an AR intelligent assistant system based on deep learning technology.Since the system can recognize the electric equipment automatically,it can help the employees to understand the conception,the operating principle and the operation method of the electric equipment better.It also can provide an assistant environment,which combines the void and the solid together.Experimental results show that this system improve the training efficiency greatly.