Revolutionizing rice cultivation: Panicle-Cloud's AI-driven approach to enhancing yield prediction and selection

Rice (Oryza sativa)  is a staple food for many people, but rice production is challenged by climate change, making it critical to improve yield traits such as panicle number per unit area (PNpM²). Recent research has focused on using computer vision and AI to quantify PNpM², with methods  such as deep learning proving effective in small-scale trials. However, these techniques face  challenges of scalability, diversity of rice varieties, and lack of large, high-quality training datasets. Overcoming these challenges is critical to developing robust, large-scale phenotypic analysis tools to improve rice production in a rapidly changing climate.

In October 2023, Plant Phenomics published a database/software article entitled “Panicle-Cloud: An Open and AI-Powered Cloud Computing Platform for Quantifying Rice Panicles from Drone-Collected Imagery to Enable the Classification of Yield Production in Rice”.

The research presents the Panicle-Cloud platform, a significant advancement in agricultural technology that provides an AI-powered cloud computing solution for quantifying rice panicles from drone-collected imagery.  The project first developed an open expert-annotated diverse rice panicle detection (DRPD) dataset, and then integrated several deep learning (DL) models into the platform. Through an iterative improvement process, the Panicle AI model emerged as the preferred choice, demonstrating superior panicle detection accuracy. To determine optimal conditions for panicle phenotyping, drone flights at different altitudes and key growth stages were analyzed, and it was found that  an altitude of 7m during early grain filling stages provided the most accurate results. Correlation analysis between Panicle-AI-derived scoring and manual counts confirmed the model's effectiveness, particularly at the 7m height, with a high correlation coefficient (R² = 0.945). The Panicle-AI model outperformed 13 state-of-the-art DL models in panicle detection accuracy. The Panicle-Cloud platform was then designed to be user-friendly, allowing non-experts to select from different AI models for panicle detection using a simple web-based interface. Users can process images individually or in batches, and the platform optimizes computation by cropping larger images. In a two-season rice field trial, the platform's ability to classify yield performance based on the PNpM² trait was tested. Using a supervised machine learning model, specifically the CatBoost algorithm, the platform successfully classified rice yield production into low, medium, and high categories with an overall accuracy of 84.03%. This feature allows rice breeders to effectively screen and select preferred varieties based on predicted yield performance.

In conclusion, Panicle-Cloud demonstrates the potential of integrating AI, cloud computing and drone technology to revolutionize rice breeding and cultivation. The platform not only improves the efficiency and accuracy of quantifying yield-related traits but also makes advanced phenotyping tools more accessible to a wider range of users, thereby increasing the ability to select high-yielding varieties in the face of global food demand challenges.

###

References

Authors

Zixuan  Teng^1,6†, Jiawei  Chen^2†, Jian Wang^3†, Shuixiu  Wu^1,4,  Riqing  Chen¹, Yaohai  Lin¹, Liyan  Shen², Robert  Jackson⁵,  Ji  Zhou^2,5*, and Changcai  Yang^1,7*

Affiliations

¹Digital  Fujian  Research  Institute  of  Big  Data  for  Agriculture  and  Forestry,  College  of  Computer  and  Information Sciences, Fujian Agriculture and Forestry University, Fuzhou 350002, China.

²State Key Laboratory  of  Crop  Genetics  &  Germplasm  Enhancement,  academy  for  Advanced  Interdisciplinary  Studies, Nanjing Agricultural University, Nanjing 210095, China.

³Ningxia Academy of Agriculture and Forestry Sciences, Yinchuan 750002, China.

⁴College of Mechanical and Electrical Engineering, Fujian Agriculture  and  Forestry  University,  Fuzhou  350002,  China.  

⁵Cambridge  Crop  Research,  National  Institute of Agricultural Botany (NIAB), Cambridge CB3 0LE, UK.

⁶Key Laboratory of Smart Agriculture and Forestry (Fujian Agriculture and Forestry University), Fujian Province University, Fuzhou 350002, China.

⁷Center for Agroforestry Mega Data Science, School of Future Technology, Fujian Agriculture and Forestry University, Fuzhou 350002, China.

About Ji  Zhou & Changcai  Yang

Ji Zhou: He is a Distinguished Professor at the Cross Research Center for Crop Phenomics, Nanjing Agricultural University.He is mainly engaged in the research of full-life span rice and wheat phenomics, and has led and developed multi-omics data fusion analysis technology based on artificial intelligence algorithms by combining computer vision, field remote sensing, and machine learning to construct multi-scale crop phenotype collection technology and core analysis algorithms.

Changcai Yang: He is an Associate Professor and the M.S. Supervisor with the College of Computer and Information Science, Fujian Agriculture and Forestry University, Fuzhou, China. He has authored or coauthored more than 60 articles. His research interests include computer vision, image processing, and point set registration.