NANJING AGRICULTURAL UNIVERSITY THE ACADEMY OF SCIENCE

The global database of the rice canopy image and corresponding rough grain yield.

The increasing global demand for staple crops, projected to rise by 60% by 2050, is a pressing issue intensified by population growth, income increases, and biofuel usage. To meet this demand sustainably, intensifying existing cropland to minimize environmental impacts and yield gaps is critical. Traditional methods like self-reporting and crop cutting are either inaccurate or impractical, and while remote sensing technologies offer potential, their application remains limited in these regions. The advent of machine learning, specifically , deep learning with convolutional neural networks (CNNs), presents a promising avenue. This technology has shown proficiency in image analysis, yet its application in diverse agricultural contexts, especially for versatile crop yield estimation across various environments and cultivars, is unexplored.

In July 2023, Plant Phenomics published a research article titled “Deep learning enables instant and versatile estimation of rice yield using ground-based RGB images “.

This research utilized a multinational database of 4,820 harvested rice plots from 7 countries, encompassing 22,067 images. These images, capturing diverse rice production systems, cultivars, and management practices, were linked to corresponding rough and filled grain yields and aboveground dry weight. Notably, in this study a Convolutional Neural Network (CNN) model was developed to estimate rough grain yield from these images. The CNN structure featured multiple convolutional layers, pooling layers, and activation functions, including ReLU, ELU, and LeakyReLU. This optimized model, after testing different learning rates and batch sizes, explained 69% and 68% of yield variation for validation and test data, respectively. The model’s accuracy was further affirmed by its close match to the observed yields, with cultivars having more data in the training set showing less deviation in yield estimation. The model’s robustness was tested under various conditions: different shooting angles, varying times of day, and dates during the ripening stage. A crucial aspect of the study was understanding the interpretation of images by the CNN model for yield estimation. Techniques like occlusion-based visualization highlighted the importance of panicles in yield estimation. This was further corroborated by a panicle removal experiment, which showed a gradual decrease in estimated yield with the removal of panicles. The model’s practical applications are manifold, offering rapid, accurate yield estimation compared to traditional estimation methods. This is particularly relevant for high-throughput phenotyping and on-station agronomic experiments. However, the study acknowledges limitations, such as the model’s reduced accuracy with lower image resolutions and its current dataset not encompassing various stress conditions.

In conclusion, this study represents a significant advancement in using ground-based RGB images for accurate rice yield prediction. It opens avenues for efficient field management, informed policy decisions, and high-throughput phenotyping, potentially revolutionizing agricultural practices and sustainability.

###

References

Authors

Yu  Tanaka^1,2*†, Tomoya  Watanabe³, Keisuke  Katsura⁴, Yasuhiro  Tsujimoto⁵, Toshiyuki  Takai⁵, Takashi Sonam Tashi Tanaka^6,7, Kensuke  Kawamura⁵, Hiroki  Saito⁸, Koki  Homma⁹, Salifou Goube  Mairoua¹⁰, Kokou  Ahouanton¹⁰, Ali Ibrahim¹¹, Kalimuthu  Senthilkumar¹², Vimal Kumar  Semwal¹³,  Eduardo Jose Graterol  Matute¹⁴, Edgar  Corredor¹⁴, Raafat El-Namaky¹⁵, Norvie  Manigbas¹⁶, Eduardo Jimmy P.  Quilang¹⁶, Yu  Iwahashi¹,  Kota  Nakajima¹, Eisuke  Takeuchi¹, and Kazuki  Saito^5,10,17*†

Affiliations

¹Graduate  School  of  Agriculture,  Kyoto  University,  Kitashirakawa  Oiwake-chou,  Sakyo-ku,  Kyoto  606-8502,  Japan.

²Graduate  School  of  Environmental,  Life,  Natural  Science  and  Technology,  Okayama  University,  1-1-1,  Tsushima Naka, Okayama 700-8530, Japan.

³Graduate School of Mathematics, Kyushu University, 744, Motooka, Fukuoka Shi Nishi Ku, Fukuoka 819-0395, Japan.

⁴Graduate School of Agriculture, Tokyo University of Agriculture and  Technology,  3-5-8  Saiwaicho,  Fuchu,  Tokyo  183-8509,  Japan.  

⁵Japan  International  Research  Center  for  Agricultural Sciences, 1-1 Ohwashi, Tsukuba, Ibaraki 305-8686, Japan.

⁶Faculty of Applied Biological Sciences, Gifu University, 1-1 Yanagido, Gifu 501-1193, Japan.

⁷Artificial Intelligence Advanced Research Center, Gifu University, 1-1  Yanagido,  Gifu  501-1193,  Japan.  

⁸Tropical  Agriculture  Research  Front,  Japan  International  Research  Center  for Agricultural Sciences, 1091-1 Maezato, Ishigaki, Okinawa 907-0002, Japan.

⁹Graduate School of Agricultural Science,  Tohoku  University,  Aramaki  Aza-Aoba,  Aoba,  Sendai,  Miyagi  980-8572,  Japan.  

¹⁰Africa  Rice  Center  (AfricaRice), 01 BP 2551 Bouaké, Côte d’Ivoire.

¹¹Africa Rice Center (AfricaRice), Regional Station for the Sahel, B.P. 96, Saint-Louis, Senegal.

¹²Africa Rice Center (AfricaRice), P.O. Box 1690, Ampandrianomby, Antananarivo, Madagascar.

¹³Africa  Rice  Center  (AfricaRice),  Nigeria  Station,  c/o  IITA,  PMB  5320,  Ibadan,  Nigeria.  

¹⁴Latin American Fund for Irrigated Rice – The Alliance of Bioversity International and CIAT, Km 17 Recta Cali-Palmira, C.P.  763537,  A.A.  6713,  Cali,  Colombia.  

¹⁵Rice  Research  and  Training  Center,  Field  Crops  Research  Institute,  ARC, Giza, Egypt.

¹⁶Philippine  Rice  Research  Institute  (PhilRice),  Maligaya,  Science  City  of  Muñoz,  3119  Nueva  Ecija, Philippines.

¹⁷International Rice Research Institute (IRRI), DAPO Box 7777, Metro Manila 1301, Philippines.

About Yu Tanaka & Kazuki Saito

Yu Tanaka: He is an associate professor at the Graduate  School  of  Agriculture, Kyoto  University.

Kazuki Saito: He is a Principal Scientist at Africa Rice Center (AfricaRice). He has made significant contributions to rice breeding projects in Africa by discovering new breeding materials, developing selection methods, and conducting other research to improve food self-sufficiency in Africa. He was also recognized for his significant achievements in improving and consolidating rice farming systems for small-scale farmers and improving farmers’ livelihoods and nutrition.

Published Date: December 20, 2023