Abstract:
The use of High Volume Instrument (HVI) system has enabled fast and reliable measurements of
cotton fiber characteristics thus producing high dimensional data. This calls for the use of
clustering techniques to adequately interpret and utilize the data. Clustering techniques classify
objects based on attributes into distinct classes (clusters). The HVI characteristics can be used to
group cotton bales so that the within group variations are kept at a minimum. This will ensure that
all the bales in a given group have the highest level of similarity hence help reduce lot to lot
variations in the manufactured yarn. A bale classification model using K-means clustering
technique and Kohonen self organizing maps (SOM) is discussed. The model is used to classify
2421 cotton bales whose HVI data containing 13 cotton attributes, was obtained from Shanghai
inspection center of industrial products and raw materials. The model reduced the 2421x13 HVI
high dimensional data into 18x13 grids, with a quantization error of 1.879 and a topographic error
of 0.083, and resulted in the identification of 16 groups of cotton bales and one group of outliers.
The outliers could be further subdivided into five subsets.