Research On Edge Detection Method Based On GPN Radial Basis Neural Network

Edge detection technology is important research directions in the field of image analyzing and processing.It is also the basic technology of many other research contents such as image compression,image segmentation,intelligent traffic management systems,target detection.Some existing edge detection methods still have some shortcomings under the effect of various interference components.For example,the edge connectivity in the detection results is not high and there are many false edges.Although edge detection method based on neural networks can improve the problem of edge connectivity,it also has the problem of low detection efficiency.For this reason,this paper proposes a new edge detection method based on GPN(Gaussian Positive-Negative)radial basis neural network.Edge detection generally includes image denoising,image enhancement,detectio n and edge positioning.The main contributions of this paper in the research and imple mentation of edge detection are as follows:(1)In the phras of image denoising,this paper introduces the thermal diffusion id ea of diffusion equation for image denoising,and makes a change to the blurring edge defect of diffusion equation,so that it can protect the edge data of original image well under the premise of iterative denoising.(2)In the subsequent edge detection stage,this paper proposes two hypotheses ba sed on the Mach effect of human visual characteristics.Based on these two assumptio ns,a new GPN contour highlighting algorithm is proposed.(3)A novel edge detection method based on GPN radial basis neural network is p roposed based on GPN contour highlighting algorithm.This method sets the weight b etween the input layer and the hidden layer of the GPN radial basis neural network str ucture to 1,because the data at the input of the neural network is pretreatment data,w hich already contains the preliminary edge information of the image,so this The settin g is to make full use of the information obtained by the pretreatment to reduce the trai ning time of some connection weights of the network,thereby improving the operatio n efficiency of the neural network.(4)When using the GPN radial basis neural network for testing,the input data ca n be directly input to the hidden layer for calculation,which avoids the mutual calcula tion between them and improves the efficiency as a whole.Finally,this paper carried out corresponding numerical experiments on the two aspects of detection effect and efficiency.For noise images and partial gray uneven images,compared with the pulse coupled neural network model,genetic neural network model and convolutional neural network model,the model in this paper has improved efficiency and better edge connectivity.The experimental results prove that the edge detection method based on GPN neural network proposed in this paper is a new and effective edge detection method,which is more efficient than the traditional neural network edge detection method and has improved detection results.