Pengembangan sistem pemutuan berbasis pengolahan citra dan jaringan syaraf tiruan untuk alat sortasi kopi beras tipe konveyor sabuk

Abstract

The objective of this research was to develop a grading system which consisted of a computer program of image processing and artificial neural network to identify quality of green coffee namely A, B, C, and RJ (reject). Two images of the green coffee from top and bottom captured by two cameras were analyzed to get six quality parameters which matched the green coffee quality criteria namely length, area, perimeter, defect area, index of red color, and index of green color. Those six quality parameters were used as training data inputs (75% - 768 kernel) of the developed artificial neural network (ANN). Eighteen variations of ANN were developed for ANN training purposes. The weight of the selected ANN architecture were used to identify the quality class of testing data (25%), and then integrated with image processing program so the program could identify green coffee quality class automatically. The total accuracy of ANN was 67% from top camera with A 59%, B 53%, C 70%, RJ 84%, and the total accuracy was 71% from bottom camera with A 75%, B 45%, C 73%, and RJ 92% based from 256 testing data. The total accuracy of ANN from combination of both cameras was 68%. New training was developed in order to increase prediction accuracy by modified ANN inputs. Prediction accuracy produced by the new ANN weight were A 78%, B 53%, C 70%, RJ 98%, and the total accuracy was 75%