关键词: 图像处理, 显著性检测, 缺陷分类, 孪生支持向量机, 密度函数, K近邻

Abstract: To enhance the classification accuracy of steel plate surface defects,a classification model with selectively weakened samples was proposed.The salient object detection algorithm was introduced in the image preprocessing step to reduce the effect of distortion in the image after binarization.To reduce the influence of unfavorable edge sample points on the model while increasing the contribution of favorable edge sample points to the model,a new density fuzzy membership function was constructed to assign weights to the samples.Based on Twin Support Vector Machine (TWSVM),the nearest neighbor density assisted fuzzy optimized TWSVM algorithm was proposed by embedding the constructed density fuzzy membership function within the model as an optimization condition to improve the classification effect.Experimental results on the dataset NEU showed that the redesigned features improved the overall accuracy by 1.66% after the introduction of the salient object detection algorithm,while the defect classification using the optimized algorithm achieved an accuracy of 98.33%,further improving the classification performance.

Key words: image processing, saliency detection, image processing, twin support vector machine, density function, K-nearest neighbor