Research on foreign body detection for granular material based on feature weighted OCSVM of color recognition

摘要

In granular materials processing, how to efficiently recognize and remove foreign bodies is very important. Training data including H, I mean and I standard deviation of target material are distinguished to be foreign bodies or not by training model, which is established using FWOCSVM method, while taking into account the characteristics of foreign body detection. A way to introduce the weight value reflecting the importance degree of attribute with its mean square deviation is developed to solve the problem, that is, attribute weights are not considered in OCSVM. The results show that FWOCSVM has more excellent performance than that of threshold or OCSVM. And color features adopted in the paper have excellent identification performance in above foreign bodies detection.