基于深度置信网络的缝纫平整度客观评价模型

Objective evaluation model of sewing flatness based on deep belief network

针对织物缝纫平整度进行客观自动评估时评价精度不高的问题,提出一种基于特征参数与深度置信网络(deep belief network,DBN)的织物缝纫平整度客观评价模型。首先,对织物缝纫样本图像进行图像灰度化处理、中值滤波去噪、直方图均衡强化等图像预处理,提取织物的纹理特征,获取更高质量、利于后续处理的织物特征图像;其次,构建灰度共生矩阵,并在0°、45°、90°与135°下提取织物图像能量、熵、对比度、相关性4个关键特征参数,在此基础上构建基于DBN的织物缝纫平整度自动评价模型,并使用缝纫图像对该模型进行训练;最后,通过提取的织物缝纫图像进行验证。实验结果显示该模型的评价精度达98.74%,与多元线性回归模型和基于BP网络模型2种方法相比,提出的评价方法可以有效对缝纫平整度进行客观评价,为织物服装外观质量控制提供理论依据。

In view of the low accuracy of objective automatic evaluation of fabric sewing flatness,an objective evaluation model of fabric sewing flatness based on feature parameters and deep belief network(DBN)is proposed.Firstly,image preprocessing including image grayscale processing,median filter denoising,and histogram equalization enhancement was performed on the image of the fabric sewing sample to extract the texture features of the fabric,so as to obtain a fabric feature image with higher quality and facilitate subsequent processing.Then the gray level co-occurrence matrix was constructed,and the four key feature parameters of energy,entropy,contrast and correlation of fabric image were extracted at 0°,45°,90°and 135°.On this basis,an automatic evaluation model of fabric sewing flatness based on DBN was constructed,and the model was trained using sewing images.Last it was verified by the extracted fabric sewing images.The results show that the evaluation accuracy of the model reaches 98.74%.Compared with the two methods of multiple linear regression model and BP network model,the proposed evaluation method can effectively evaluate the flatness of sewing objectively and provide a theoretical basis for the quality control of fabric garment appearance.