基于改进DenseNet201网络的织物疵点检测算法

Fabric Defect Detection Algorithm Based on Improved DenseNet201 Network

针对当前织物疵点检测存在大多数采用人工检测、速度慢、检测准确率低等问题,提出一种改进DenseNet201网络的织物检测算法。先对数据集图像进行预处理,可视化各种织物疵点类型的数量,把数据集划分为正常织物、8种常见织物疵点,对疵点图像进行数据增强,从而扩增训练集数量;然后提取在数据集ImageNet下预训练好的DenseNet201权重参数进行迁移学习,改进卷积层第1层、添加SPP层和本研究9分类的分类层;最后经过反复调整参数训练得到织物疵点检测模型。试验表明:改进后的DenseNet201模型对正常织物、8种常见织物疵点识别精度为96.8%。认为:改进DenseNet201网络模型具有良好的泛化性和鲁棒性。

Aiming at the problems of manual detection,lower speed and accuracy in fabric defect detection,an improved fabric detection algorithm of DenseNet201 network was proposed.Firstly,the data set was preprocessed to visualize the number of various fabric defect types.The data set was divided into normal fabric and 8 common defects.The defect images were enhanced to amplify the number of training sets.Then,the pre-trained DenseNet201 weight parameter migration learning under the ImageNet data set was extracted to improve the first layer of convolution layer,adding SPP layer and the classification layer of the 9-classification in this paper.Finally,the fabric defect detection model were obtained after adjusting the parameter training repeatedly.The results showed that the improved DenseNet201 model could identify normal fabric,8 common defects with an accuracy of 96.8%.It is considered that the improved DenseNet201 network model has better generalization and robustness.