基于多层感知器与特征组合的皮革缺陷分类

Classification of Leather Defects based on Multilayer Perceptron and Feature Combination

针对皮革表面存在不均匀纹理的特性、缺陷分类难度大的问题,构造了一种多层感知器结合特征组合的分类方法。首先通过Gabor滤波方法强调皮革表面纹理特征,并获取皮革图像的5个灰度分布统计特征与8个灰度共生矩阵统计特征,以此描述皮革表面纹理和缺陷的特征;然后采用多层感知器结合特征组合的分类方法对皮革缺陷图像进行分类。实验验证结果表明,该方法对四类皮革样本的分类准确率可达98%以上。

In order to solve the problem of uneven texture on leather surface and difficulty in defect classification,a classification method based on Multilayer Perceptron combined with feature combination was constructed.Firstly,the texture features of leather surface were emphasized by Gabor filtering method,and 5 gray level distribution statistical features and 8 gray-level co-occurrence matrix statistical features of leather image were obtained to describe the texture and defect characteristics of leather surface.Then,Multilayer Perceptron combined with feature combination was used to classify leather defect images.The experimental results show that the classification accuracy of the method can reach more than 98%.