Material Defect Identification System Based On Improved Support Vector Machine

With the development of communication technology and machine learning technology,material defect detection technology based on Channel State Information(CSI)has gradually entered the public’s vision.Currently,fine-grained CSI can be obtained on inexpensive Wi Fi devices,and action recognition is the classification of different action features,providing relatively more environmental information and making it relatively easy to distinguish.However,for material defects,they are often subtle defects.Compared to action recognition,material defect detection provides less fine-grained information and is much more difficult to distinguish.In response to the issue of limited environmental information provided by material defect detection,this paper conducts research on a material defect detection system based on improved Support Vector Machine(SVM)and CSI.The main work achievements of the paper include:(1)This article proposes using Differential Evolution(DE)algorithm to optimize SVM.In the face of nonlinear data problems in practical application scenarios,we choose to use SVM based on radial basis function(RBF)kernel function for classification and recognition.The classification performance of SVM using RBF kernel function is affected by regularization parameters and RBF kernel function parameters,which directly determine the recognition accuracy and learning ability of nonlinear SVM.However,there is currently no conclusive theoretical basis to guide the selection of these two parameters,and common grid cross validation search methods often cannot find the optimal combination of parameters.The performance of DE proposed in this article in parameter optimization is superior to traditional grid cross validation methods,and it can effectively optimize the parameters of SVM to obtain the optimal training model.(2)This article proposes a material defect detection system based on improved SVM and CSI.Due to the use of cheap Wi Fi devices for data collection in this system,it is not possible to provide sufficient environmental information for classification when the tested object is stationary.Therefore,a method of uniformly moving the tested object between the sending and receiving ends is proposed for data collection,enabling it to obtain sufficient environmental information for classification.The system mainly utilizes CSI collected from inexpensive Wi Fi devices to extract environmental information related to material defects.First,the amplitude outlier are removed through the Hampel filter,and then the linear interpolation algorithm is used to ensure the correctness of the data format.Then,principal component analysis(PCA)and Butterworth filter are used to jointly denoise the data,and then PCA is used to extract the characteristics.Through sample training,the performance of the DE-SVM model is optimized,and finally,the preprocessed dataset is classified through DE-SVM.This article takes building materials as an example,and the experimental results show that the DE SVM proposed in this article has performance advantages over standard SVM in material defect detection,and can achieve a higher true positive rate.