Date Published: May 29, 2019
Publisher: Public Library of Science
Author(s): Kyosuke Yamamoto, Rongling Wu.
Convolutional neural networks (CNNs) can not only classify images but can also generate key features, e.g., the Google neural network that learned to identify cats by simply watching YouTube videos, for the classification. In this paper, crop models are distilled by CNN to evaluate the ability of deep learning to identify the plant physiology knowledge behind such crop models simply by learning. Due to difficulty in collecting big data on crop growth, a crop model was used to generate datasets. The generated datasets were fed into CNN for distillation of the crop model. The models trained by CNN were evaluated by the visualization of saliency maps. In this study, three saliency maps were calculated using all datasets (case 1) and using datasets with spikelet sterility due to either high temperature at anthesis (case 2) or cool summer damage (case 3). The results of case 1 indicated that CNN determined the developmental index of paddy rice, which was implemented in the crop model, simply by learning. Moreover, CNN identified the important individual environmental factors affecting the grain yield. Although CNN had no prior knowledge of spikelet sterility, cases 2 and 3 indicated that CNN realized about paddy rice becoming sensitive to daily mean and maximum temperatures during specific periods. Such deep learning approaches can be used to accelerate the understanding of crop models and make the models more portable. Moreover, the results indicated that CNN can be used to develop new plant physiology theories simply by learning.
Recently, machine learning has experienced tremendous advancements. Deep learning has provided solutions to many tasks that could not be solved by conventional machine learning. One remarkable achievement of deep learning is AlphaGo  (developed by DeepMind), a computer program that plays the game Go and can beat professional human Go players without any handicaps.
Herein, distillation of crop models was conducted to investigate the ability of deep learning to find the plant physiology knowledge behind the distilled crop model from given data. Although most research utilizing machine learning in agriculture lacks sufficient data, this problem was overcome via simulations using crop models. Interestingly, the performance of the model generated by the distillation was analyzed. In addition, the learnings obtained by the model and determination of whether there were any cues related to plant physiological theories behind the distilled crop model were analyzed. CNN, a state-of-the-art method based on deep learning, was used for distillation.