Fruit Detection for Classification by Type with YNOv3-Based CNN Algorithm

Abstract

The fruit is part of the flowers in plants that are produced from pollination of the pistils and stamens. The shape and color of many fruits with a variety, with the type of single fruit, double fruit and compound fruit. This study asks for the development of 10 pieces detection applications to help the sensor agriculture sector for 10 pieces detection. The data in this study used the image of 10 fruits namely Mangosteen, Delicious, Star Fruit, Water Guava, Kiwi, Pear, Pineapple, Salak, Dragon Fruit, and Strawberry. Training and testing using CNN algorithms and YOLOv3 machine learning methods with the support of the work of the Darknet53 neural network. The analysis was conducted using 2,333 images of data from 10 classes. The training process is carried out up to 5,000 iterations stored in checkpoints. The implementation of the detection of 10 pieces was carried out on Google Collaboratory through imagery with two tests. Accuracy in the detection of 10 pieces can reach more than 90% in the first test of each fruit and an average of 70% in the second test for images outside the test data.

Keywords

• detection
• yolov3(you only lock once)
• cnn(convolutional neural network)
• darknet
• google colaboratory