A cans waste classification system based on RGB images using different distances of k-means clustering
Abstract. This study aims to build a classify the cans waste based on the pixel of captured Red, Green, and Blue (RGB) image by implement different metric 3 distances of k-means clustering; Manhattan, Euclidean, and Minkowski metric distance. The image capturing is designed using combinations of two the conveyor belt speeds of 0.181 m/sec and 0.086 m/sec, two the lightings of halogen and incandescent lamps, and four lighting angles of 300, 450, 600, and 900. The classification results note that the implementation of Manhattan distance on the k-means clustering method for classifying the cans waste into three can types has the highest level of accuracy in the majority of data. The highest accuracy level of classification is obtained from data of captured image on the conveyor belt speeds of 0.181 m/sec, the lightings of halogen lamp, and the lighting angles of 450 by implementing the Euclidean distance, while the lowest accuracy level of classification is obtained from data of captured image on the lighting angles of 300 with the same speeds and the lamp by implementing the Manhattan distance. The highest average accuracy is obtained by implementing the Euclidean distance, that derived from the average accuracy at lighting angle of 450.