Efficient detection method for foreign fibers in cotton

Abstract

Since foreign fibers in cotton seriously affect the quality of the final cotton textile products, machine-vision-based detection systems for foreign fibers in cotton are receiving extensive attention in industrial equipment. As one of the key components in detection systems, the suitable and good classifier is significantly important for machine-vision-based on detection systems for foreign fibers in cotton due to it improving the system’s performance. In the study, we test five classifiers in the dataset of foreign fibers in cotton, and for finding the best feature set corresponding to the classifiers, we use the four filter feature selection approaches to find the best feature sets of foreign fibers in cotton corresponding to specific classifiers. The experimental results show that the extreme learning machine and kernel support vector machines have the excellent performance for foreign fiber detection and the classification accuracy are respectively 93.61% and 93.17% using the selected corresponding feature set with 42 and 52 features.