Recognizing Of Parts Based On The Combined Invariant Moments And Optimized SVM

With the rapid development of modern manufacturing industry,the production scale of manufacturing enterprises is expanding and the productivity is increasing.The original method of using human eyes to sort the parts in the production line is usually of low efficiency and high error rate.With the upgrading of the industry,the manual sorting method can't meet the application requirements.Therefore,based on machine vision,this thesis designs an intelligent recognition method of parts images which combines combined invariant moments and optimized Support Vector Machine(simply,SVM).This method consists of two very important parts: feature vector construction and image classification and recognition.Firstly,by comparing Hu invariant moments,Affine invariant moments and Zernike invariant moments,it can be concluded that constructing the feature vector with single invariant moments will decrease the image description information and lower the accuracy of recognition.Therefore,the mentioned three kinds of invariant moments are combined to construct the feature vector.Then,the feature vector is associated with the parts imags' class label to construct the training set and test set.Firefly algorithm and particle swarm optimization algorithm are used to optimize the relevant parameters of SVM,and the parts images are classified and identified by the optimized SVM.In the process of parameter optimization,the recognition accuracy of training set is regarded as the objective function value,and the penalty factor and kernel function parameter are obtained to initialize the SVM training model.In the process of recognition,the test set is the input of the trained SVM,and its output value is the class label of the parts image.Finally,through comparative experiments,it proves that the recognition method based on combined invariant moments and SVM optimized by Firefly Algorithm(simply,FA-SVM)performs better in recognition accuracy and stability.And this research has certain theoretical and application value for other recognition work based on machine vision.