半监督FSVM在羽绒菱节识别中的应用

摘要

Currently, most work of feather and down category recognition is done by man with a microscope, but this method has many disadvantages. FSVM based on Partial Supervision (PS-FSVM) is applied to feather and down category recognition. PS-FSVM is used to increase the size of training samples, which is based on a small number of labeled training samples. Then it uses the characteristics of FSVM to reduce the impact of misclassified samples which caused by semi-supervised learning. After the image processing, the triangle node of two-value image of feather is to be recognized with PS-FSVM. The results show PS-FSVM improve the recognition rate of the triangle node.%目前,我国对羽绒种类的识别主要由人工借助于显微镜完成,这种方法存在许多不足.提出将半监督FSVM算法引入到羽绒识别中,用半监督学习方法以少量的训练样本为基础,扩大训练样本集的规模,同时利用FSVM的特性减少半监督学习所带来的误差;利用半监督FSVM对经过处理的羽绒二值化图像中的菱节进行识别.该方法提高了菱节识别的准确率.