A multiple instance learning approach to study leaf wilt in soybean plants

Authors: BANERJEE, SANJANA - North Carolina State University; RAMOS, PAULA - North Carolina State University; REBERG-HORTON, CHRIS - North Carolina State University; Mirsky, Steven; Locke, Anna; LOBATON, EDGAR - North Carolina State University

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/7/2025
Publication Date: 3/13/2025
Citation: Banerjee, S., Ramos, P., Reberg-Horton, C.S., Mirsky, S.B., Locke, A.M., Lobaton, E. 2025. A multiple instance learning approach to study leaf wilt in soybean plants. Computers and Electronics in Agriculture. 15(6):614. https://doi.org/10.3390/agriculture15060614.
DOI: https://doi.org/10.3390/agriculture15060614

Interpretive Summary: oybean leaf wilting is an obvious drought stress indicator that can be used by soybean breeders to select drought tolerant varieties or by growers to evaluate how much water their crop needs. While effective, leaf wilting ratings are slow and labor intensive to collect. Automated ratings could improve efficiency in soybean breeding programs or enable remote sensing of crop stress for management purposes. Here, we developed a machine learning method, Multiple Instance Learning, to automatically evaluate an image of a soybean plot and assign a rating that indicates the degree of wilting. This method is an improvement over previous algorithms, with 64% perfect ratings and 94% almost-perfect ratings when compared with human expert-rated images. Additionally, the model provided more consistent drought ratings than human experts when asked to re-rate a previously rated image. This research will benefit farmers by making it faster and easier to evaluate soybeans for drought stress, which is useful both for breeding more drought-resistant varieties and for managing existing varieties.

Technical Abstract: Recent years have seen significant technological advancements in precision farming and plant phenotyping. Remote sensing along with deep learning (DL) techniques can increase phenotyping efficiency and help on-farm decision making with rapid stress detection. In this work, we use these techniques to evaluate drought stress in soybean plants, a crop whose yield is significantly affected by water availability. Images were taken from a high vantage in the field at various times throughout the day. Each image is given a wilting score ranging from 0 to 4 by expert scorers. We implement a DL method called multiple instance learning (MIL) to perform wilt classification as well as generate heat maps that highlight wilt levels in specific regions of the image. Given the significant overlap between adjacent classes in our dataset, we were able to achieve an overall classification accuracy of 64% and a one-off accuracy of 96% on our holdout test set. Our model outperformed DenseNet121 in most metrics, and provided comparable performance to a vision transformer (ViT) while having fewer parameters overall, less complexity (useful for edge implementations), and some interpretability. Furthermore, we were able to show that our model outperformed expert human annotators by predicting more consistent and accurate wilt levels when considering single-image re-annotation. The results show that our proposed methodology can be a useful approach in detecting drought stress in soybean fields to facilitate efficient crop management and aid selection of drought-resilient varieties.