Aiming at the phenomenon that the adhesion of conductive particles seriously affects the detection accuracy of thin film transistor liquid crystal display(TFT-LCD)bumps,a binary classification image detection method for the conductivity of bumps with convolutional neural network(CNN)incorporating K-means clustering feature selection was proposed.A binary classification model of qualified and unqualified bump image detection was established based on the standard of judging whether the number of particles attached to the bump was qualified or not.A data set containing 2 000 bump images was constructed according to the supervised learning mechanism.In order to improve the detection accuracy,further aiming at the characteristics of differences in the gray threshold between conductive particles and the background in different bump images,a CNN bump image particle number classification algorithm incorporating K-means clustering feature selection was proposed.A 10-fold cross-validation method was applied to evaluate the effectiveness of the algorithm,and the algorithm accuracy rate reached 96.0％,which was 8％higher than that of Otsu's method and 5.5％higher than that of the watershed method.
关键词
薄膜晶体管液晶显示器/质量检测/卷积神经网络/K均值聚类
Key words
thin film transistor liquid crystal display/quality inspection/CNN/K-means clustering
维普
万方数据