Development of Navigation System for Tea Field Machine Using Semantic Segmentation
Journal
IFAC-PapersOnLine
Journal Volume
52
Journal Issue
30
Pages
108-113
Date Issued
2019
Author(s)
Lin Y.-K
Abstract
Labor shortage is a critical issue in most of industries, especially in agricultural production. In recent year, riding-Type tea plucking machine was imported to provide a relatively high-efficient solution for tea harvesting. However, high-level driving skill is essential. Improper operation may cause damage on tea trees and also lead to mechanical failure. A real-Time image-based navigation system may provide an automatic choice to mitigate the difficulties. In this study, deep neural network architectures were applied to semantic segmentation to derive the contours of the tea rows and identify the obstacles in the field scene. Performance of four models including 8s-, 16s-, 32s-of the fully convolutional networks (FCN) and ENet were compared. Considering the overall performance, ENet outperformed other models with the mean intersection over unit (mean IU) of 0.734, the mean accuracy of 0.941, and the inference time of 0.176 s. Furthermore, Hough transform was introduced to obtain the guidelines based on the classification. The average bias of angles and distance were 6.208° and 13.875 pixels, respectively. The preliminary result showed the feasibility of using the developed navigation system for field application. To achieve higher precision, images that cover a diverse scenario in the field were being collected and trained in future work. © 2019 Elsevier B.V. All rights reserved.
Subjects
Automatic navigation; deep learning; image processing; semantic segmentation; tea plucking machine
SDGs
Other Subjects
Agriculture; Deep learning; Deep neural networks; Failure (mechanical); Hough transforms; Image processing; Navigation systems; Network architecture; Neural networks; Semantics; Tea; Agricultural productions; Automatic navigation; Convolutional networks; Critical issues; Field application; Mechanical failures; Real time images; Semantic segmentation; Image segmentation
Type
conference paper