[6-1130-P] Other Categories (6th)

[6-1130-P-13] Classification of Sugarcane Variety using Image Processing and Multivariate Analysis

*KITTIPON APARATANA¹, Hiroo Takaragawa^1,2, Yoshinari Izumikawa^1,2, Eizo Taira¹ (1. Faculty of agriculture, University of the Ryukyus, Okinawa 903-0213(Japan), 2. The United Graduate School of Agricultural Sciences, Kagoshima University, Kagoshima 890-0065(Japan))

キーワード：Sugarcane, Variety classification, PCA, PLS-DA, SVM-C

Sugarcane variety classification is essential for data collecting and learning for the breeder. It is difficult for a farmer to identify a sugarcane variety without specialist help. In this research, three Japanese sugarcane varieties (Ni15, Ni22, and Ni27) from six areas in the south of Japan were classified according to full pixel and color features of the sugarcane bud. The 54 images of sugarcane bud were acquired from the sugarcane field using a mobile phone’s digital camera, equipped with a fixed distance accessory. To develop classification models, two types of data; The full pixel data and color feature data from images were investigated for input to the model. The full pixel and color features were subjected to Principal component analysis (PCA) to describe the sugarcane bud samples. Then, the samples were classified into varieties by performing partial least squares discriminant analysis (PLS-DA) and support vector machine classification (SVM-C). The results of the full pixel show that the pooled classification rates (averaged three classification rate) by PLS-DA and SVM-C were 79.6% and 84.5%, respectively, while the pooled classification rates by PLS-DA and SVM-C of the color features were 75.9% and 74.1%, respectively. Therefore, these results show that the size and color spaces of sugarcane buds could be the keys to classifying sugarcane varieties and that the best way of classifying Japanese sugarcane (Ni15, Ni22, and Ni27) was the SVM-C method using full pixel of sugarcane bud.