Rice provides 20% of the world’s food energy, with expected demand rising by 60% by 2050, but climate change, more extreme weather and evolving pathogens threaten farmers’ ability to produce enough of the crop. To help better predict crop yield — and how interventions may improve yield — an international research team has developed an approach that farmers may be able to implement using just a smartphone.
The researchers published their method on July 28 in Plant Phenomics.
“Accurately assessing crop productivity in developing countries is not easy,” said co-author Keisuke Katsura, associate professor in the Graduate School of Agriculture at Tokyo University of Agriculture and Technology. “Our main concern was to develop technology that could be easily used by anyone in the agriculture field. Many African farmers now own smartphones, so we wanted to develop technology that could be used with them.”
According to Katsura, there are three well-known approaches for assessing crop yield — self-reporting, crop cutting and remote sensing. Self-reporting is often inaccurate; crop cutting is costly in terms of time and labor as it involves physically harvesting some of the crop; and remote sensing requires specialized tools and knowledge.
“The absence of reliable data on agricultural statistics is a serious constraint for both agricultural research and policy,” Katsura said. “There is need to monitor agricultural productivity and evaluate the impact of productivity-enhancing interventions, without labor intensive crop cuts or knowledge intensive remote sensing technologies.”
The team collected more than 22,000 digital images of rice canopies in harvesting plots across Côte d’Ivoire, Senegal, Japan, Kenya, Madagascar, Nigeria and Tanzania. The photos were taken about three feet above the canopies, at various angles and in different lighting during the rice’s ripening stage and during harvest. The researchers also collected quantitative data on each crop’s yield, feeding that information and the images into a convolutional neural network (CNN) model. Considered a deep-learning-based model, the CNN contains several layers to optimize what it learns from the data. For example, it can sort through the images and identify which ones include signs of rice maturity, so it only needs to examine those images for further information, instead of the entire dataset.
“This is the first study to develop a versatile CNN model to predict rice yield accurately only be using ground-based images captured via digital camera,” Katsura said, explaining that previous attempts to apply the CNN model were limited to specific growing environments. “Our model was able to rapidly and effectively estimate rice yield at a low cost with satisfactory precision in the existing most comprehensive and international dataset in terms of growing environment and management practices, number of cultivars, camera angles and time of days.”
The model instantaneously predicted crop yield with a 68% accuracy, which the researchers said is comparable or even higher than predictions made in earlier studies using satellite data or in combination with other data. When the researchers reduced the resolution of the images, the model still predicted crop outcome with nearly 60% accuracy.
“The CNN structure used in this study is much smaller than what was used in previous studies — implying that the developed model can be easily transferred to mobile devices such as smartphones or scaled by the use of unmanned aerial vehicles,” Katsura said.
The researchers said next, they plan to investigate how their method might help diagnose nutrient deficiencies, pests and other problems.
“We’re also interested in applying this technology to other crops,” Katsura said. “Yield studies are widely conducted in agriculture experiment stations everywhere in the world. There is much that can be done by making better use of such data.”
Other collaborators include co-corresponding author Yu Tanaka, Yu Iwahashi, Koto Nakajima and Eisuke Takeuchi, Graduate School of Agriculture, Kyoto University; Tomoya Watanabe, Graduate School of Mathematics, Kyushu University; Yashuhiro Tsujimoto, Toshiyuki Takahiro, Kensuke Kawamura and co-corresponding author Kazuki Saito, Japan International Research Center for Agricultural Sciences; Takeshita Sonam Tashi Tanaka, Faculty of Applied Biological Sciences, Gifu University; Hirohito Saito, Tropical Agriculture Research Front, Japan International Research Center for Agricultural Sciences; Koki Houma, Graduate School of Agricultural Science, Tohoku University; Salifou Goube Mairoua and Kokou Ahouanton, Africa Rice Center, Côte d’Ivoire; Ali Ibrahim, Africa Rice Center, Regional Station for the Sahel, Senegal; Kalimuthu Senthilkumar, Africa Rice Center, Madagascar; Vimal Kumamoto Semwal, Africa Rice Center, Nigeria; Eduardo Jose Graterol Matute and Edgar Corredor, Latin American Fund for Irrigated Rice — The Alliance of Bioversity International and CIAT; Raafat El-Namaky, Rice Research and Training Center, Field Crops Research Institute, Egypt; and Norvie Manigbas and Eduard Jippy P. Quilting, Philippine Rice Research Institute, Philippines. Yu Tanaka is also affiliated with Okayama University’s Graduate School of Environmental, Life, Natural Science and Technology. Takashi Sonam Tashi Tanaka is also affiliated with Gifu University’s Artificial Intelligence Advanced Research Center. Saito is also affiliated with the Africa Rice Center, Côte d’Ivoire, and the International Rice Research Institute, Philippines.
The European Union, the International Fund for Agricultural Development, CGIAR Research Program, Japan Society for the Promotion of Science, and JICA/JST SATREPS supported this research.
For more information about the Laboratory of Crop Production Science, Tokyo University of Agriculture and Technology, please visit https://sites.google.com/site/tuatcroplabeng/home.
About Tokyo University of Agriculture and Technology (TUAT):
TUAT is a distinguished university in Japan dedicated to science and technology. TUAT focuses on agriculture and engineering that form the foundation of industry, and promotes education and research fields that incorporate them. Boasting a history of over 140 years since our founding in 1874, TUAT continues to boldly take on new challenges and steadily promote fields. With high ethics, TUAT fulfills social responsibility in the capacity of transmitting science and technology information towards the construction of a sustainable society where both human beings and nature can thrive in a symbiotic relationship. For more information, please visit http://www.tuat.ac.jp/en/.
Deep Learning Enables Instant and Versatile Estimation of Rice Yield Using Ground-Based RGB Images
Article Publication Date