Research On Machine Vision System With Image Target Recognition Based On Structural Balance Network

The primary task of a machine vision system is to measure and judge the object,so that the system can recognize the object,locate and track it.Among them,image recognition is an important method to assist the machine vision system to complete identification and positioning tasks.At present,image recognition algorithms have achieved many valuable results.There are many existing methods of image recognition,each of which shows its advantages in a specific field.These algorithms are filling their advantages under their specific conditions and are widely used in industrial and social fields,such as unmanned driving,intelligent transportation,eye in the sky,drones,and smart medical.However,Existing image recognition methods also have some problems.For instance,a wide range of alternative models of image recognition algorithms based on convolutional neural networks in the field of image recognition.The application usually requires high hardware requirements(generally requires the configuration of GPU),and the brakes adapt to a large Storage and calculation of training samples.In addition,there are gradient dissipation and gradient explosion problems in the network training process,which will affect the high speed and high recognition rate of the recognition method.Therefore,it is worth further studying other image recognition methods that maintain fast recognition speed and high accuracy.Over the last decade,image recognition methods based on complex networks have evolved into a distinctive research direction in the field of image recognition research.A complex network can be considered as a coupling of node groups and dynamical edges.The topology of the network is only related to the connection relationship between the nodes,and has nothing to do with the order and location of the nodes.Therefore,in theory,the introduction of complex network methods into image recognition can greatly reduce the impact of image changes(stretching,rotation,translation,scaling,etc.)on image recognition.However,such methods also have certain drawbacks:(1).The connection relationship formed by the gray value difference(or the distance metric)between the nodes is not appropriate,and the resulting of the image Characteristic face cannot guarantee a high recognition rate;(2).The pixels are regarded as the nodes of the network,and the increase of nodes cannot guarantee high speed.In view of the above problems,some scholars have proposed improved image recognition methods(hybrid methods)for complex networks.The core idea is to reduce the number of nodes in the network,which plays a certain role in increasing the speed of the recognition algorithm.However,the improved recognition method still directly uses the pixels of the image as the nodes of the network.When the number of nodes increases,it will affect the high speed of the image recognition algorithm.In view of the problems listed above,how to make appropriate improvements and extensions constitutes the motivation and starting point of the research in this article.Based on this,this paper first proposes an image recognition method based on structural balance network.To this end,this paper first proposes an image recognition way based on a structural balance network,revisit gray level image from a new perspective,the image grayscale matrix is treated as the structural balance network connection relationship matrix,Hadamard product transformation is used to classify the connection weights to obtain image recognition feature parameters,thereby improving the speed and accuracy of image recognition.On this basis,by changing the network topology,that is,selecting different Hadamard transformation matrices to traverse the grayscale image,and then using the method of weighting the feature parameters,the recognition speed and accuracy are improved.Through actual simulation experiments,compare with the existing many commonly used image recognition methods from the recognition accuracy and speed,the main advantages of the image recognition algorithm given in this paper are: For the gray matrix of the image,the gray value of the image is treat as the connection relationship weight of the structural balance network,and the identification parameters of some topological features of the scale network are extracted.The identification parameters are significantly reduced.Compared with commonly used image recognition algorithms,recognition speed and accuracy are improved,and try to use the machine vision system.