摘要
针对传统图像特征提取方法用于缺陷检测存在自适应程度低的问题,提出了一种基于拓扑独立成分分析(TICA)的无监督图像特征提取缺陷识别方法。首先通过TICA算法从缺陷集中自适应地估计基向量,利用基向量对应滤波器与缺陷图像进行滤波,提取滤波响应作为特征,为避免TICA算法陷入局部最优,引入差分进化(DE)算法进行优化。然后采用ReliefF算法和K-means算法对提取特征进行选择,减少特征中冗余和无关信息,降低特征向量维数。最后利用随机森林分类器对样本进行缺陷分类,目前总体识别准确率高达96.0%,验证了所提出方法的有效性。
Aiming at the problem of low self-adaptation of traditional image feature extraction methods for defect detection,an unsupervised image feature extraction defect recognition method based on topological independent component analysis(TICA)is proposed.First,use the TICA algorithm to adaptively estimate the basis vector from the defect set,use the filter corresponding to the basis vector to filter the defect image,and extract the filter response as a feature.In order to avoid the TICA algorithm from falling into the local optimum,the differential evolution(DE)algorithm was introduced for the optimization.Then the ReliefF algorithm and K-means algorithm were used to select the extracted features,reduce the redundancy and irrelevant information in the features,and reduce the dimensionality of the feature vector.Finally,the random forest classifier was used to classify the samples.The current overall recognition accuracy is as high as 96.0%,which verifies the effectiveness of the proposed method.
作者
宋晓宇
冯加华
袁帅
陈智丽
SONG Xiao-yu;FENG Jia-hua;YUAN Shuai;CHEN Zhi-li(School of Information and Control Engineering,Shenyang Jianzhu University,Shenyang Liaoning 110168,China)
出处
《计算机仿真》
北大核心
2023年第7期94-99,125,共7页
Computer Simulation
基金
国家自然科学基金(62073227)。
关键词
拓扑独立成分分析
差分进化
特征提取
Topological independent component analysis
Differential evolution
Feature extraction