Research On Image Recognition Techniques Of Decorated Granite Based On Color Features And Logistic Regression

With the continuous development of our country's stone industry,more and more stone images have emerged.Facing these massive and disorganized stone image resources,how to identify and classify them has become a difficult problem for the stone industry.In recent years,content-based image recognition(CBIR)technology has been commonly used for such research and has achieved very good results.Considering that the main stones are granite and marble in the market,due to the advantages of granite which have high hardness,hard weathering,and color appearance that can remain unchanged for more than a hundred years,the subject of granite is used as a research object for identification,and color features is used for the image recognition of decorated granite.The research contents of this paper includes:1.Image identification of ornament granite based on multi-two-classification logistic regression algorithm.This paper proposes two aspects of the research program:(1)In this study,the original training set is improved by the down-sampling method and the up-sampling method based on the smote algorithm.The experiment shows that the up-sampling method based on the smote algorithm can improve the training set,and improve the accuracy rate of granite recognition.(2)Under the condition of scheme one,this paper proposes a color feature extraction method in the HSV color space that performs non equal interval quantization and equal interval quantization.Experiments show that the use of color feature extraction methods with non equal interval quantization can improve the accuracy rate of granite recognition.2.Image identification of ornament granite based on multi-classification logistic regression algorithm.There are three aspects of the research program:(1)In the HSV color space,the color feature extraction methods with non equal interval quantization and equal interval quantization are respectively performed to explore the effect of two color feature extraction methods on granite image recognition.Experiments show that the use of color feature extraction method with non equal interval quantization can improve the accuracy of granite recognition.(2)Under the condition of scheme one,this paper reduces the dimension of the color features by Principal Component Analysis(PCA),and explores the effect of granite image recognition before and after dimension reduction.Through experiments,it has been shown that using PCA to reduce the dimension of features can improve the overall accuracy rate of granite identification.(3)Under the condition of scheme two,the classifier is improved,by combining the logistic regression,random forest and support vector machine,to explore the influence of granite image recognition before and after improvement.The results show that the improved classifier can improve the accuracy rate of granite recognition.3.Finally,the two above-mentioned large improvement schemes are compared experimentally,and a scheme that is more conducive to improving the image recognition of ornament granite is selected.Namely,it is a multi-two-classification research scheme that selects the up-sampling method based on the smote algorithm in the training set improvement scheme and selects a non equal interval quantization feature extraction method based on the HSV color space in the color feature selection.