https://scholars.lib.ntu.edu.tw/handle/123456789/605950
Title: | Developing a guiding and growth status monitoring system for riding-type tea plucking machine using fully convolutional networks | Authors: | Lin Y.-K Chen S.-F YAN-FU KUO Liu T.-L Lee S.-Y. |
Keywords: | Automatic navigation;Deep learning;Fully convolutional network (FCN);Semantic segmentation;Tea plucking machine;Convolutional neural networks;Harvesting;Plants (botany);Semantic Segmentation;Semantic Web;Semantics;Convolutional networks;Fully convolutional network;Plantation managements;Status monitoring;Tea plantations;Tea plants;Convolution;artificial nest;artificial neural network;crop plant;image processing;machinery;monitoring system;tea;Camellia sinensis | Issue Date: | 2021 | Journal Volume: | 191 | Source: | Computers and Electronics in Agriculture | Abstract: | Tea (Camellia sinensis) is one of the most popular beverage crops worldwide. As tea production is labor-intensive, riding-type tea plucking machines have been employed to improve harvest efficiency. High level of operating skills are essential to maneuver these machines. Improper operation may cause damage to tea plants and reduce the quality of harvested tea leaves. Besides harvest, monitoring the growth status of tea canopies is another essential task of tea plantation management. Monitoring conducted through conventional manual observation is time consuming. Therefore, this study attempted to solve these two issues by developing a system that guided an operator to steer a riding-type tea plucking machine in real-time and provided growth status of tea plants after each scouting trip. Fully convolutional networks (FCNs) and image processing techniques were applied to develop the guiding subsystem. This study compared two FCNs, namely FCN-8 s and ENet. The results showed that ENet outperformed FCN-8 s in generating a guiding line for the operator. Statistical analysis indicated that, with the guiding subsystem, the deviation between the driving direction and tea row center was significantly reduced. The guiding subsystem also could detect obstacles on the tea row where it was operated on. When in operation, the riding-type tea plucking machine was used as a platform to collect images of tea canopies for tea plantation management. The monitoring subsystem provided a graphical user interface consisting of a route recording, a growth status map, a canopy image of tea plants, and growth status ratings of the tea plants. FCN-8 s performed better than ENet in quantifying the growth status of tea plants, with a mean IU of 0.964 and a mean accuracy of 0.989. ? 2021 Elsevier B.V. |
URI: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85119198534&doi=10.1016%2fj.compag.2021.106540&partnerID=40&md5=02c438bf0b546f47d1d924bb40032e46 https://scholars.lib.ntu.edu.tw/handle/123456789/605950 |
ISSN: | 01681699 | DOI: | 10.1016/j.compag.2021.106540 |
Appears in Collections: | 生物機電工程學系 |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.