摘要
为对光学薄膜缺陷图像进行准确识别分类,提出一种基于改进的卷积神经网络光学薄膜缺陷图像识别方法。为突出输入图像中的缺陷信息,采用改进的LBP算法对图像进行预处理。从三个方面对传统的卷积神经网络进行改进:为了解决单通道卷积神经网络对图像特征提取不充分的问题,构建双通道卷积神经网络;改进传统的ReLU激活函数,避免模型出现欠拟合现象;使用支持向量机(SVM)代替Softmax分类器,提高计算效率和准确率。光学薄膜缺陷图像仿真识别实验表明,所提方法分类平均准确率高达93.2%,训练时间为964 s,充分验证了所提方法的鲁棒性和有效性。
In order to accurately identify and classify the defect images of optical films,an optical film defect image identification method based on improved convolution neural network is proposed.To highlight the defect information in the input image,the improved LBP algorithm was used to preprocess the image.The traditional convolution neural network was improved from three aspects:construct a two-channel convolution neural network to solve the problem of insufficient image feature extraction by single-channel;improve the traditional ReLU activation function to avoid under-fitting of the model;use SVM instead of Softmax classifier to improve computational efficiency and accuracy.The simulation recognition experiment of optical film defect images shows that the average classification accuracy of the proposed method is as high as 93.2%,and the training time is 964 s,which fully verifies the robustness and effectiveness of the proposed method.
作者
张振华
陆金桂
李浩然
Zhang Zhenhua;Lu Jingui;Li Haoran(School of Mechanical and Power Engineering,Nanjing Tech University,Nanjing 211816,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2021年第4期197-203,共7页
Computer Applications and Software
关键词
光学薄膜
缺陷识别
改进的LBP
改进激活函数
双通道
支持向量机
卷积神经网络
Optical film
Defect identification
Improved LBP
Improved activation function
Two-channel
Support vector machine
Convolutional neural network