Clustering Analysis for Cotton Trash Classification

Abstract

Raw cotton may contain various kinds of trash, such as leaf, bark, and seed coat particles. The content of each of these trash categories is useful information for finding more efficient cleaning processes and predicting the quality of the finished products. This paper addresses the importance of using chromatic and geometric features of trash for trash description, and presents three different clustering methods that automatically classify trash based on the feature measurements. Compared with the geometric attributes of trash, such as size and shape, color attributes are less changeable during harvesting and ginning of cotton and are therefore more reliable and descriptive in categorizing trash. Three clustering methods—sum of squares, fuzzy, and neural network—prove effective for trash classification. Sum of squares clustering and fuzzy clustering require iterative computations and generate comparable classification accuracy. Neural network clustering yields the highest accuracy, but it needs more computational time for network training.