Japan Geoscience Union Meeting 2022

Presentation information

[J] Poster

M (Multidisciplinary and Interdisciplinary) » M-TT Technology & Techniques

[M-TT46] Introducing metaverse to agriculture. Are we ready?

Fri. Jun 3, 2022 11:00 AM - 1:00 PM Online Poster Zoom Room (35) (Ch.35)

convener:Seishi Ninomiya(Graduate School of Agriculture and Life Sciences, the University of Tokyo), convener:Yukihiro Takahashi(Department of Cosmosciences, Graduate School of Science, Hokkaido University), Chairperson:Seishi Ninomiya(Graduate School of Agriculture and Life Sciences, the University of Tokyo)

11:00 AM - 1:00 PM

[MTT46-P03] Diagnosis of Garlic health status from reflectance spectral data using machine learning

*Shuichi Ando1, Yusuke Iida1, Yukihiro Takahashi2, Nobuyasu Naruse3, Yurino Ishida2 (1.Niigata University, 2.Hokkaido University, 3.Shiga University of Medical Science)


Vegetation index using data acquired by satellites is required to improve agricultural efficiency. NDVI is calculated from the reflectance of two wavelength bands and has been used traditionally as a vegetation index. However, it has been reported that NDVI is affected by some environmental factors, and it is desired to develop a noble vegetation index which can be used in wider cases and with better accuracy. To this end, we developed machine learning models for diagnosis of garlic health status using wide range reflectance spectral data and compared the accuracy of them and that using NDVI.
For the reflectance spectral data of garlic acquired by simple devices, we developed four-class classification model (healthy, rust, leaf blight, and soft rot) and binary classification model (healthy and diseased). We used three machine learning algorithms: Neural Network (NN), Random Forest (RF), and Support Vector Machine (SVM). We achieved the highest accuracy of 0.80. The accuracy of binary classification is better than four-class classification for all machine learning methods. On the other hand, the model training was unstable due to the small amount of data. Therefore, we conducted data augmentation for each health state and performed a model training with the augmented data set. Moreover, we compared the accuracy of them with that using NDVI. As a result, binary classification accuracy by NDVI is 0.71, which is comparable to that of the machine learning method. On the other hand, NDVI has a large overlap of distributions for each health state, which leads that it is difficult to develop four-class classification model, while the machine learning model achieved four-class classification accuracy of 0.80. This indicates that the combination of wide range spectra and machine learning model can provide more accurate and detailed classification than NDVI only.