Publication: Machine Learning Methods for Assessing Freshness in Hydroponic Produce
Issued Date
2018-07-02
Resource Type
Other identifier(s)
2-s2.0-85065089779
Rights
Mahidol University
Rights Holder(s)
SCOPUS
Bibliographic Citation
2018 International Joint Symposium on Artificial Intelligence and Natural Language Processing, iSAI-NLP 2018 - Proceedings. (2018)
Suggested Citation
Konlakorn Wongpatikaseree, Narit Hnoohom, Sumeth Yuenyong Machine Learning Methods for Assessing Freshness in Hydroponic Produce. 2018 International Joint Symposium on Artificial Intelligence and Natural Language Processing, iSAI-NLP 2018 - Proceedings. (2018). doi:10.1109/iSAI-NLP.2018.8692883 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/45614
Research Projects
Organizational Units
Authors
Journal Issue
Thesis
Title
Machine Learning Methods for Assessing Freshness in Hydroponic Produce
Other Contributor(s)
Abstract
© 2018 IEEE. Smart farms are increasing in both number and level of technology used. Image processing had been applied to hydroponic farms to detect disease in plants, but detecting the freshness of vegetable had not been addressed as much. In this work we applied image processing and machine learning technologies to the task of distinguishing between fresh and withered vegetable. We compared 3 classical machine learning classifier: decision tree, Naive Bayes, Multi-Layer Perceptron; and one type of deep neural network. Manual feature extraction was performed for the classical machine learning, while the input to the deep neural network was the raw images. We collected the data by taking one image of the vegetable every 10 minutes for one week each time. We labeled the data by considering vegetable from day 1 and day 2 to be fresh while from day 3 onward was considered wither. Experiment results show that the best model for this task was decision tree with a test accuracy of 98.12%. Deep neural network did not perform as well as expected. We hypothesize that the reason is due to overfitting of the training data since the training accuracy for deep neural network was as high or even higher than other classifiers.