Implementasi Convolutional Neural Network Pada Alat Klasifikasi Kematangan dan Ukuran Buah Nanas Berbasis Android

Abstract

Abstrack - South Sumatra is the region with the highest production of pineapples in 2021. The process of selling pineapples depends on the size and maturity. Farmers classify pineapples subjectively with both eyes, causing the classification process to be ineffective. Machine learning technology is developing very rapidly, one of which is deep learning which uses very deep neural networks to learn feature representations of data automatically. This study aims to implement the Convolutional Neural Network (CNN) algorithm to classify the ripeness and size of pineapples so that the sorting process for pineapple production can be effective and accurate. There are 6 classification labels, namely, large ripe pineapple, large half ripe, medium ripe, medium half ripe, small ripe, and small half ripe. Raspberry Pi 3B+ and Pi camera are used as fruit image capture tools. The results of the training process accuracy were 99.4%, and the validation process accuracy was 92.4% with a dataset of 275 data for each label. The dataset is used 80% as training data and 20% as validation data. Meanwhile, for testing the tool, 90 test data were used with an accuracy of 90.83%. And the results of the classification will appear on the Android application including the amount of pineapple stock that has been detected, so that it can make it easier for farmers to sort pineapples.