Mangosteen Detection Using Deep Learning

Abstract

Robot technology has developed more and more rapidly. Robot were developed to harvest fruit instead of humans. Robot uses a camera and  computer to  process images,  detect fruit and location on branches  to control a mechanical arm to pick fruit. Image detection technique is therefore an important technology use to development of vision robot technology. Thailand's mangosteen export is one of the top in the world. A Mangosteen is a tall trunk which makes it difficult to harvest, to detect mangosteen on plants, must be classify is young or ripe, lies on  the branches and may be covered by the leaves. This research presents the detection of mangosteen on plants by image using Convolutional Neural Network (CNN) compare with Faster R-CNN deep learning framework. The research methods are as follows 1) image preparation 2) training deep learning to mature or ripe mangosteen detection 3) testing to   mature or ripe mangosteen detection. The performance of detection model employs the confusion matrix. The results show that can mature or ripe mangosteen detection, the model of CNN is an accuracy of 50.62%, precision of 26.42%, recall of 93.33% and F1-Score of  41.18%, the Faster R-CNN is an accuracy of 90.22%, precision of 63.16%, recall of 48.98% and F1-Score of 55.17%.