Tomato Growth Stage Monitoring for Smart Farm Using Deep Transfer Learning with Machine Learning-based Maturity Grading
Tomato Growth Stage Monitoring for Smart Farm Using Deep Transfer Learning with Machine Learning-based Maturity Grading
复制标题
DOI：
10.17503/agrivita.v42i1.2499
复制
发表时间：
2020-02-01
期刊：
影响因子：
0.8
通讯作者：
Vicerra, Ryan Rhay P.
中科院分区：
文献类型：
--
作者：
de Luna, Robert G.;Dadios, Elmer P.;Vicerra, Ryan Rhay P.
番茄种植业需要在应用该技术方面采取新的思路，以便对其生长进行监测和监控。机器视觉和图像处理技术在日益增长的水果质量检测需求中变得非常有用，特别是西红柿。本文介绍了一种计算机视觉监测系统的设计和开发，该系统通过检测花和果实的存在来评估室内番茄植株的生长情况。该系统还提供番茄果实的成熟度分级。研究中使用了两种预训练的深度迁移学习模型，即基于区域的卷积神经网络（R-CNN）和单镜头检测器（Single Shot Detector）。采用人工神经网络（ANN）、k近邻（KNN）和支持向量机（SVM）对番茄果实进行成熟度分类。评价结果表明，对于花和水果的检测，R-CNN对花和水果的检测总体准确率分别为1.67%和19.48%，而SSD对花和水果的检测总体准确率分别为100%和95.99%。在成熟度分级的机器学习中，SVM的训练测试准确率为97.78% ~ 99.81%，KNN为93.78% ~ 99.32%，ANN为91.33% ~ 99.32%。
The tomato farming industry needs to adopt new ideas in applying the technology for its growth monitoring and main. Machine vision and image processing techniques have become useful in the increasing need for quality inspection of fruits, particularly, tomatoes. This paper deals with the design and development of a computer-vision monitoring system to assess the growth of tomato plants in a chamber by detecting the presence of flowers and fruits. The system also provides maturity grading for the tomato fruit. Two pre-trained deep transfer learning models were used in the study for the detection of flowers and fruits, namely, the Regional-based Convolutional Neural Network (R-CNN) and the Single Shot Detector (SDD). Maturity classification of tomato fruits are implemented using the Artificial Neural Network (ANN), K-Nearest Neighbors (KNN), and the Support Vector Machine (SVM). Evaluation results show that for the detection of flowers and fruits, the over-all accuracy of the R-CNN is 1.67% for flower detection and 19.48% for the fruit detection while SSD registered 100% and 95.99% for flower and fruit detection respectively. In the machine learning for maturity grading, SVM produced the training-testing accuracy rate of 97.78%-99.81%, KNN with 93.78%-99.32%, and ANN with 91.33%-99.32%.