Research On Glass Products Sorting Method Based On Feature Fusion

Glass products are daily products with large market demand.Classified recycling and recycling of some commonly used glass products can reduce the supply pressure of glass products,reduce the resource consumption and environmental pollution in the manufacturing process.So,glass products sorting can save resources and protect the environment.However,due to the large quantity,various shapes and sizes of recycled glass products,the efficiency of manual sorting is low and the error rate is high.Therefore,this paper proposes a method of glass products sorting based on feature fusion by using target detection technology.The main research contents of this sorting method are as follows:A selective search algorithm is used to obtain the candidate regions of glass products.However,there is a large time cost in the calculation of texture similarity.In this paper,HLSN texture feature is proposed to improve the selective search algorithm.In the process of merging different image regions to form candidate regions,the integral graph algorithm is introduced to accelerate the HLSN features corresponding to different image regions,so as to reduce the calculation times of texture features corresponding to overlapping parts between different image regions.Experimental results show that the improved selective search algorithm has a good performance in the generation of candidate regions.On the basis of obtaining the candidate regions of glass products,this paper proposes the FIW-CCA feature fusion algorithm to form the fusion features corresponding to the candidate regions.Firstly,the features of candidate regions are extracted by using the PHOG and CLBP algorithms.Then,the features of PHOG and CLBP corresponding to the training set are processed by using the FIW algorithm to form the SFIW feature set.Finally,the CCA model with the maximum correlation strategy is introduced to eliminate the redundant information from features,and the SFIW feature set generated is used to solve the CCA model to generate the FIW-CCA feature fusion model.Experiments show that the fusion features formed by the FIW-CCA mode have strong expression ability.Because of the weak generalization ability and low classification accuracy when using a single classifier to classify the fused features,this paper uses stacking integrated classifier to classify the fused features of FIW-CCA.SVM,KNN,MNB and LightGBM are used as the base classifier and meta classifier respectively.In order to prevent over fitting,the cross validation algorithm is used to train the three base classifiers,and the output probability vector of the base classifier is used as the input data of the meta classifier,so as to enhance the generalization ability of the stacking integrated classifier.The experimental results show that the proposed sorting method is effective.