A computer vision method to characterize the types of coffee beans based on color and texture analysis Yusuf Hendrawan (a*), Bagas Rohmatulloh (a), Muchammad Riza Fauzy (b), Retno Damayanti (a), Mochamad Bagus Hermanto (a), Dimas Firmanda Al Riza (a), Sandra (a)
a) Laboratory of Mechatronics and Agro-industrial Machineries, Department of Agricultural Engineering, Universitas Brawijaya, Jl. Veteran, Malang, ZIP 65145, Indonesia
*yusuf_h[at]ub.ac.id
b) Department of Industrial Engineering, Universitas Merdeka, Jl. Terusan Raya Dieng 62-64, Malang, ZIP 65146, Indonesia
Abstract
Several types of local Indonesian coffee have been recognized in the international market. The gap in coffee prices triggers the occurrence of counterfeiting of famous coffee products. Therefore, it is necessary to develop a non-destructive system that can recognize the external appearance characteristics of each type of coffee bean. This study aimed to develop a computer vision system to characterize three types of Indonesian Arabica coffee beans i.e. Gayo Aceh, Kintamani Bali, and Toraja Tongkonan based on the external appearances. Each type of coffee bean was analyzed for its characteristics based on color and textural features. Color features included the average value of red, green, blue, grey, hue, saturation(HSL), saturation(HSV), lightness, value, X(XYZ), Y(XYZ), Z(XYZ), C(CMY), M(CMY), Y(CMY), C(CMYK), M(CMYK), Y(CMYK), K(CMYK), L(Lab), a(Lab), b(Lab), C(LCH), H(LCH), U(LUV), and V(LUV). Textural features included energy, entropy, contrast, homogeneity, inverse difference moment, correlation, sum-mean, variance, cluster tendency, and maximum probability on every type of color space. From the features extraction, a total of 286 types of image features were analyzed. The results showed that 19 features of 26 color features and 185 features of 260 textural features had the potential to characterize and classify three types of coffee beans. Of the total 286 image features, the three image features which were recommended to have the best performance with the smallest and most stable standard deviations i.e. X(XYZ) sum-mean (average standard deviation 0.01186), Y(XYZ) sum-mean (average standard deviation 0.01187), and Z(XYZ) sum-mean (standard deviation 0.01419).
