Logo Tropentag

Tropentag, September 14 - 16, 2022, Prague

"Can agroecological farming feed the world? Farmers' and academia's views."

Deep learning-based estimation of rice yield using RGB image

Kazuki Saito1, Yu Tanaka2, Tomoya Watanabe3, Keisuke Katsura4, Yasuhiro Tsujimoto5, Toshiyuki Takai5, Takashi Tanaka6, Kensuke Kawamura5, Hiroki Saito4, Koki Homma7, Salifou Goube1, Kokou Ahouanton1, Ali Ibrahim8, Kalimuthu Senthilkumar9, Vimal Semwal10, Yu Iwahashi2, Kota Nakajima2, Eisuke Takeuchi2

1Africa Rice Center, Cote d'Ivoire
2Kyoto University, Japan
3Kyushu University, Japan
4Tokyo University of Agriculture and Technology, Japan
5Japan International Research Center for Agricultural Sciences, Japan
6Gifu University, Japan
7Tohoku University, Japan
8Africa Rice Center, Senegal
9Africa Rice Center, Madagascar
10Africa Rice Center, Nigeria


Crop productivity is poorly assessed globally. The absence of reliable data on agriculture statistics is a serious constraint for both agricultural research and policy. The objective of this study is to develop a deep learning-based approach for estimating rice yield using RGB images. During ripening stage and at harvest, over 22,000 digital images were captured vertically downwards over the rice canopy from a distance of 0.8 to 0.9 m across 20 locations in seven countries. Rice yields were obtained in the corresponding area ranging from 0.1 and 16.1 t ha-1. A convolutional neural network (CNN) applied to these data at harvest predicted 70% variation in yield with a relative root mean square error (rRMSE) of 0.22. Furthermore, the accuracy of CNN model was evaluated using data which were not used for model development. The result showed that the model successfully detected cultivar difference in yield, and yield variation associated with water management practices. The robustness of the CNN model to image quality was further tested using the images taken (i) from different shooting angles, (ii) at various times of day during the five days before harvest, and (iii) on different shooting dates during the ripening stage. The CNN demonstrated robustness against the images acquired at different shooting angles with 30 degree angle from right angle, in diverse light environments, and during late ripening stage. Our work suggests that this low-cost, hands-on, and rapid approach can provide a breakthrough solution to assess the impact of productivity-enhancing interventions and identify fields where these are needed to sustainably increase crop production.

Keywords: Convolutional neural network, RGB image, rice, yield

Contact Address: Kazuki Saito, Africa Rice Center, Abidjan, Cote d'Ivoire, e-mail: k.saito@cgiar.org

Valid HTML 3.2!