Lasdun Violet, Güereña Davíd, Ortiz-Crespo Berta, Mutuvi Stephen, Selvaraj Michael, Assefa Teshale
London School of Economics, Houghton St., London WC2A 2AE, United Kingdom.
Alliance of Bioversity and the International Center for Tropical Agriculture (CIAT), TARI - Selian, Dodoma Road, Arusha, Tanzania.
Agric Syst. 2024 Oct;220:104054. doi: 10.1016/j.agsy.2024.104054.
Crop breeding in the Global South faces a 'phenotyping bottleneck' due to reliance on manual visual phenotyping, which is both error-prone and challenging to scale across multiple environments, inhibiting selection of germplasm adapted to farmer production environments. This limitation impedes rapid varietal turnover, crucial for maintaining high yields and food security under climate change. Low adoption of improved varieties results from a top-down system in which farmers have been more passive recipients than active participants in varietal development.
A new suite of research at the Alliance of Bioversity and CIAT seeks to democratize crop breeding by leveraging mobile phenotyping technologies for high-quality, decentralized data collection. This approach aims to resolve the inherent limitations and inconsistencies in traditional visual phenotyping methods, allowing for more accurate and efficient crop assessment. In parallel, the research seeks to harness multimodal data on farmer preferences to better tailor variety development to meet specific production and consumption goals.
Novel mobile phenotyping tools were developed and field-tested on breeder stations in Colombia and Tanzania, and data from these trials were analyzed for quality and accuracy, and compared with traditional manual estimates and absolute ground truth data. Concurrently, Human-Centered Design (HCD) methods were applied to ensure the technology suits its context of use, and serves the nuanced requirements of breeders.
Computer vison (CV)-enabled mobile phenotyping achieved a significant reduction in scoring variation, attaining imagery-modeled trait accuracies with Pearson Correlation values between 0.88 and 0.95 with ground truth data, and reduced labor requirements with the ability to fully phenotype a breeder's plot (4 m × 3 m) in under a minute. With this technology, high-quality quantitative phenotyping data can be collected by anyone with a smartphone, expanding the potential to measure crop performance in decentralized on-farm environments and improving accuracy and speed of crop improvement on breeder stations.
Inclusive innovations in mobile phenotyping technologies and AI-supported data collection enable rapid, accurate trait assessment and actively involve farmers in variety selection, aligning breeding programs with local needs and preferences. These advancements offer a timely solution for accelerating varietal turnover to mitigate climate change impacts, while ensuring developed varieties are both high-performing and culturally relevant.
由于依赖人工视觉表型分析，全球南方的作物育种面临“表型瓶颈”，这种方法既容易出错，又难以在多个环境中进行扩展，从而阻碍了对适应农民生产环境的种质的选择。这一限制阻碍了品种的快速更新，而品种快速更新对于在气候变化下维持高产和粮食安全至关重要。改良品种的低采用率源于一种自上而下的系统，在该系统中，农民在品种开发中更多地是被动接受者，而非积极参与者。
国际生物多样性联盟和国际热带农业中心开展的一系列新研究旨在通过利用移动表型技术进行高质量、分散的数据收集，使作物育种民主化。这种方法旨在解决传统视觉表型分析方法固有的局限性和不一致性，从而实现更准确、高效的作物评估。同时，该研究旨在利用关于农民偏好的多模态数据，更好地调整品种开发以满足特定的生产和消费目标。
开发了新型移动表型分析工具，并在哥伦比亚和坦桑尼亚的育种站进行了实地测试，对这些试验的数据进行了质量和准确性分析，并与传统人工估计值和绝对地面真值数据进行了比较。同时，应用以人为本的设计（HCD）方法，以确保该技术适合其使用环境，并满足育种者的细微需求。
基于计算机视觉（CV）的移动表型分析显著减少了评分差异，图像建模的性状准确率与地面真值数据的皮尔逊相关值在0.88至0.95之间，并且降低了劳动力需求，能够在一分钟内对育种者的地块（4米×3米）进行全表型分析。借助这项技术，任何拥有智能手机的人都可以收集高质量的定量表型数据，扩大了在分散的农场环境中测量作物性能的潜力，并提高了育种站作物改良的准确性和速度。
移动表型技术和人工智能支持的数据收集方面的包容性创新能够实现快速、准确的性状评估，并让农民积极参与品种选择，使育种计划与当地需求和偏好保持一致。这些进展为加速品种更新以减轻气候变化影响提供了及时的解决方案，同时确保所培育的品种既高产又符合文化需求。
