PhenoMix

Motivation

Climate change and the increasing demand for sustainable food production call for agricultural systems that use resources more efficiently while maintaining high yields. Crop mixtures, particularly combinations of cereals and legumes, offer a promising approach by improving resource use efficiency, biodiversity, and resilience. However, understanding how different crop species and varieties interact under changing environmental conditions remains challenging, especially because traditional phenotyping methods are labor-intensive and difficult to scale. The project seeks to improve our understanding of crop mixtures and support the selection of optimal variety combinations for more sustainable agriculture.

Proposed Approach / Solution

Phenomix focuses on developing machine learning and computer vision methods to enable automated trait estimation and transfer learning for pure crop stands and crop mixtures. The approach plans to explore pretrained foundation models for plant trait extraction, alongside 3D reconstruction and image generation to create realistic synthetic plant images. Domain adaptation techniques facilitate the transfer of models trained on high-quality images from the FIP 2.0 platform (Fig. 1) to more accessible imaging devices, such as drones, while minimizing the need for additional labeled data. Furthermore, human-in-the-loop and active learning strategies incorporate expert feedback to improve model predictions and guide data annotation. The developed methods are evaluated under real field conditions and contribute to comprehensive, multimodal time-series datasets, supporting downstream analyses of crop growth and interactions (Fig. 2).

‍

Figure 1: Field phenotyping platform (FIP) with the FIP 1.0 sensor head (A), FIP 2.0 multi-view sensor head (B) and an overview of the field with experimental plots (C).

‍

Figure 2: Overview of the project. Foundation models for automated plant trait estimation (A), multi-view imaging, 3D reconstruction, and synthetic image generation for domain transfer (B), human-in-the-loop strategies and field evaluations to improve model performance (C).

Impact

PhenoMix contributes to more sustainable and climate-resilient agriculture by supporting the identification of suitable cereal–legume variety combinations that improve resource use efficiency and crop productivity. By making advanced phenotyping methods accessible beyond specialized research platforms, the project facilitates large-scale monitoring of crops under diverse field conditions and supports data-driven decisions in crop breeding and agricultural management. Beyond these practical applications, PhenoMix provides openly accessible, multimodal time-series datasets and reusable machine learning methods, enabling further advances in plant phenotyping, crop modeling, and the study of plant–environment interactions. These resources create a foundation for future research and foster collaboration between data science, plant science, and agricultural research.