Research Of Cotton Foreign Fiber Feature Processing Based On Automated Visual Inspection

Cotton industry takes an important position in the national economy of China, but the quality of cotton fiber is low because of containing lots of foreign fiber. The proportion of foreign fiber in cotton fiber, but, it will seriously affect the quality of cotton yarn and textiles, if not removed. Currently, systems of inspecting and removing foreign fibers in cotton are mainly based on automatic vision, but the inspecting and removing rate is generally about 80%.For finding the excellent feature set, lots of features are extracted from the cotton foreign fiber objects, which include color, texture, shape features. The improved ant colony algorithm is proposed and applied to feature selection, by which the optimal feature subset is obtained. Automated visual inspection, image processing, feature extracting, and pattern recognition are main techniques used in this research. The main research works are as follows:(1) Foreign fiber Image Object extracting. The 24 color features are extracting from the cotton foreign fiber objects, which come from two color spaces, RGB and HSV. The 43 texture features are extracting from the cotton foreign objects, which are described by Gray-level Co-occurrence Matrix, Gray-gradient Co-occurrence matrix, gray-smooth co-occurrence matrix and Gray-level differences. The 8 shape features are extracted, which is suitable to describe the cotton foreign fiber. The 75 dimension vector consists of the color, texture, shape features, which are extracted by above methods.(2) Feature selection for foreign fiber object. To obtain features to describe a foreign fiber object accurately, an improved ant colony algorithm (IACA) was developed to select the optimal feature subset effectively and efficiently. The initial probability pretreatment and section mutation was adopted in IACA. The initial probability pretreatment adopts the classification performance of the individual feature as the first heuristic information, which can induce the impact of the redundant features and speed up the convergence. Section mutation based on the MCFS characteristic of cotton foreign fiber, randomly adjusts the selected features on a certain percentage and ensures the subset distributed in the three groups: color, texture and shape feature, after ants selecting the features every time, as a result, this can avoid the local convergence. By this way, the IACA could obtain strong search ability at the beginning of the evolution and could achieve accelerated convergence along evolution. IACA was tested using features extracted from cotton foreign fiber objects, and was compared with the ant colony algorithm (ACA) under the same condition. The results show that IACA gets improved searching ability and accelerated convergence speed compared with IACA. The optimal feature subset selected by IACA has smaller size than that selected by ACA. This is very important for online classification of foreign fibers. (3) Design and implementation of the prototype system of feature extracting and selecting for detection of foreign fibers in lint. The objective of the prototype system is to test the above mentioned methods. The system of the prototype includes 4 parts, i.e., parameter setting, image processing, feature extracting, feature selection. The software system includes 4 modules, i.e., parameter setting, image processing, feature extracting and feature selection. Great deals of experiments indicate that the developed method works effectively and efficiently.