Detection of Unripe Tangerines Using Convolutional Neural Network for Early Yield Estimation
Taegyun Rho, Byoung-Kwan Cho
Department of Biosystems Machinery Engineering, Chungnam National University, 99 Daehak-ro, Yuseong-gu, Daejeon, 34134, South Korea
Abstract
Early yield estimation of tangerine is crucial to predict market price at harvesting time, which helps farmers to take precautions to ensure their minimum income. The conventional method for the yield estimation based on human eyes is inaccurate, subjective, and time-consuming. Hence, the demand for the development of accurate, rapid, and automated yield estimation methods has been increased. In this study, an advanced image analysis technique, convolutional neural network (CNN) was used to detect tangerines in the trees at the unripe stage. The CNN transfer learning model was designed based on ResNet-50 of which the last 10 network layers were modified to be suitable for tangerine detection. Images from 130 tangerine trees were fed into the constructed CNN model for training followed by validation. The model could distinguish unripe tangerines from green leaves, stems, and various backgrounds with an over 92% accuracy. The result demonstrates that the developed CNN model has potential to be used for the early estimation of tangerine yield.
Keywords: Yield estimation, image analysis, tangerine, artificial intelligence
Topic: Other Agricultural and biosystems topic