基于支持向量机的多特征融合纤维分类算法

Multi feature fusion fiber classification algorithm based on support vector machine

针对市面上常用的人工鉴别法无法对多种类纤维的识别分类的问题,提出了一种新的适用于多种类纤维图像识别分类的多特征融合纤维分类算法。首先提取10类纤维图像的灰度直方图、局部二值模式(LBP)、方向梯度直方图(HOG)、Hu不变矩和灰度共生矩阵(GLCM)特征,然后再将上述特征加权融合得到一个新特征,利用SVM模型对其进行训练(8000根纤维)和测试(2000根纤维),从而得到最终识别准确率。结果表明:该算法的平均准确率为85.8%,其中腈纶、醋酯纤维以及锦纶3类纤维的特征非常明显准确率达到90%以上,同时较难分辨的羊毛、羊绒纤维准确率也达到88%左右。该算法较好的达到了识别效果,为快速准确识别纤维提供技术基础。

A new multi feature fusion fiber classification algorithm suitable for multi type fiber image recognition and classification were proposed to address the problem that commonly used manual identification methods in the market cannot recognize and classify multiple types of fibers.Firstly,the grayscale histograms,local binary patterns(LBP),directional gradient histograms(HOG),Hu invariant moments,and gray level co-occurrence matrix(GLCM)features of 10 types of fiber images were extracted.Then,the above features were weighted and fused to obtain a new feature,which is trained(8000 pices fiber)and tested(2000 pices fiber)using an support vector machine(SVM)model to obtain the final recognition accuracy.The experimental results show that the average accuracy of the algorithm is 85.8%,among which the characteristics of acrylic fiber,acetate fiber,and nylon fiber are very obvious,and the accuracy reaches over 90%.At the same time,the accuracy of difficult to distinguish wool and cashmere fibers also reaches about 88%.This algorithm has achieved good recognition results,providing a technical foundation for fast and accurate identification of fibers.