GA-SVC based search applied for optimization of image features subset in quality estimation system of bulk green coffee bean

This research aims to develop a quality estimation system of bulk coffee grain based on machine vision technique that was mainly focused on finding the best subset of image feature combination. The subset was defined as the minimum number of features for achieving the reasonable level of identification or interpretation. For this purpose, a heuristic searching method based on genetic algorithm (GA) was applied to find the best feature subset from 26 image features extracted from gray channel (9-color features and 17-co-occurrence-based-textural features). The GA with binary code chromosome was designed with a support vector classifier (SVC)-based fitness function which also played as pattern recognition software for such developed-machine vision system. The experiment was started with data collection of image samples captured by a constant illumination of 200 lux of an imaging system. Besides varied the sample (7-grades for Arabica and 8-grades for Robusta), the study also evaluated some preconditioning treatments for the initial image. With a constant population of 80 chromosomes, the selection step was performed until the 20^th generation with standard genetic operations (selection, crossover, mutation, elitism), the algorithm was able to obtain an optimal feature subset consisting in average number of 5-7 features for all tested data sets. Evaluation on the analysis result shows that the best identification level was achieved from directly image processing (without preconditioning). By the pre-processing step, a quality estimation system based on selected feature subset was potentially able to estimate the quality of green coffee beans in bulk with accuracy of 86% for Arabica and 87% for Robusta coffee.