Abstract: Computer image processing technology is more and more widely used in the field of textiles and garments. The technology can detect and extract the target information required in the image, and the cross application of textile and garment promotes the development of intelligent detection technology in the production process, such as fabric defect detection and sewing flatness detection, improving the production efficiency. However, the intelligent detection technology of clothing appearance defects has developed slowly, mainly relying on experienced patternmakers to judge the types of defects, which is undoubtedly not conducive to the improvement of production efficiency. Therefore, the application of image processing technology in clothing appearance inspection, avoiding subjective accidents, enhancing the objectivity of judgment results, and reducing the demand for manpower and material resources are urgent problems for clothing enterprises.
In order to realize the automatic identification of clothing appearance defect types, we took the dress images of men〖WT《Times New Roman》〗'〖WT〗〖WT5BZ〗〖HT5SS〗s suit sleeves as the research object, and proposed a discrimination method combining image processing technology and BP neural network technology. Firstly, we analyzed the visual influencing factors of sleeve defect dress images, dissected the appearance fold characteristics of some sleeve defect types, and determined the parameters for quantifying sleeve defect folds. Then, with the help of image processing software MATLAB, we extracted the fold parameters of sleeve defect dress image samples, used the image grayscale, grayscale enhancement, image threshold segmentation set image binarization and other technologies to preprocess the dress image of the defect map, and extracted the parameters such as fold width and fold depth based on the grayscale curve chart, and on the basis of the processed binarization diagram, we extracted the slope of the fold direction of the parameters on the fold trend, and extracted the fold parameters of four types of defects in 52 samples. Finally, we wrote the program of BP neural network to identify the drawback model. 70% of the data was used as the training set and 30% was input to the BP neural network training as the test set. The input was the extracted three fold parameters, and the output was an encoded number representing the type of defects. It is verified that the model has high accuracy and stability, can identify the types of sleeve defects, and realize the automatic judgment of sleeve appearance defects.
The relationship between different sleeve defect types and their corresponding appearance folds provides enlightenment for the development of intelligent detection technology for clothing appearance quality, and the types of clothing appearance defects can be automatically determined by using image processing technology to to extract the characteristic parameters of different appearance folds and combining with the neural network model. The research results can provide reference guidance for the development of clothing appearance quality inspection technology.

Key words: men 's suit sleeves, defect type, folds, image processing, BP neural network, MATLAB

摘要： 为实现自动判别男西装袖的弊病类型，提出了一种将图像处理技术与BP神经网络相结合的判别方法。首先收集不同弊病类型的男西装袖图像，借用MATLAB平台，对图像进行灰度化、灰度增强、二值化等预处理，绘制褶皱部位的灰度曲线图；然后基于灰度曲线图以及二值化图提取褶皱宽度、褶皱深度和褶皱斜率等3个特征参数；最后将提取的特征参数和对应的弊病类型输入到BP神经网络中训练和识别，对男西装袖弊病图像的类型进行分类。结果显示,提出的方法对袖弊病类型的判别具有较高的准确率与稳定性。

关键词: 男西装袖, 弊病类型, 褶皱, 图像处理, BP神经网络