Sudden death syndrome (SDS) is an important disease in soybean (Glycine max L.) caused by the soilborne fungus Fusarium virguliforme (Bradley et al. 2021). First identified in Richland County, North Dakota in 2018, SDS has become an increasing concern. A survey of 30 commercial soybean fields conducted in 2024 in southeastern ND revealed that SDS was present in 50% of the surveyed fields, with disease incidence up to 60%. The spread of SDS was also observed in previously unreported counties, such as Cass and Dickey. Notably, above-average rainfall during the 2024 growing season resulted in wet soils conducive to the development of SDS. The causal fungus infects soybean roots early in the growing season, producing toxins that result in foliar symptoms, typically appearing during reproductive stages as chlorotic and necrotic lesions (Rodriguez et al. 2021). Detecting SDS primarily relies on observing these symptoms; however, they can be confused with those of brown stem rot (Westphal et al. 2008). Current detection methods, including field scouting, expert diagnoses, and uprooting plants to examine root rot symptoms, can be labor-intensive (Raza et al. 2020). Techniques, such as machine vision and remote sensing, offer promising alternatives for early disease diagnosis, which is crucial for effective management (Ali et al. 2019). Hyperspectral sensing, in particular, measures reflectance across a wide spectral range, enabling the identification of subtle crop condition changes and facilitating efficient screening for SDS resistance in breeding programs (Bajwa et al. 2017). In addition, hyperspectral imaging, using spectral information, can provide chemometric data on healthy and stressed plants. This chemometric data can then be utilized to train advanced deep learning (DL) models. These models are the backbone of a decision-making application that provides the user with crucial time to effectively manage soybean diseases, saving both time and economic resources.

North Dakota, along with other northern states, contributes to producing three-quarters of the total U.S. soybean production. In 2023, major soybean-producing states accounted for 76.6 percent of total U.S. soybean yield loss, while only 4.8 percent of soybean production was lost due to diseases. Soybeans are a big contributor to North Dakota’s economic development – even a slight percentage reduction in yield can translate to significant financial loss. Recent and ongoing advancements in sensor technologies and artificial intelligence have enabled the development of data-driven decision-making models at low cost. Such models can be adapted to multiple smart agriculture systems, including cloud-based applications, existing farm equipment, and agricultural robotics while forecasting the plants’ susceptibility to diseases.

Objective 1: Identify the hyperspectral regions most associated with plants infected by F. virguliforme in the greenhouse and field conditions.
Objective 2: Development of soybean SDS dataset using proximal sensors and field platforms.
Objective 3: Development of deep learning models with a focus on reducing computational cost, size of the model, and investigation of their deployment capabilities on edge devices.

A dataset of soybean plants affected by SDS disease collected using proximal hyperspectral sensor in the range of 400–1000nm. Soybean leaves images will be captured in greenhouse and field conditions.

List of significant wavelengths in the near infrared region that are most representative of SDS disease in soybean.

A well generalized DL model that can accurately detect early plant stress due to SDS disease in soybean plants.

The integration of hyperspectral sensing, machine learning, and data-driven decision-making offers significant benefits to soybean farmers in managing Sudden Death Syndrome (SDS), a growing concern caused by Fusarium virguliforme. Early detection is crucial for effective disease management, as SDS can cause substantial yield loss. Hyperspectral sensors can detect subtle spectral changes in stressed plants long before visible symptoms appear, providing farmers with valuable time to intervene. By combining hyperspectral data with machine learning models, farmers can predict disease development, optimize resource allocation, and reduce input costs by targeting areas at highest risk. This data-driven approach not only preserves crop yield and improves profitability but also supports more sustainable farming by minimizing over-application of pesticides and reducing environmental impact. The use of these technologies in conjunction with smart farming systems, including cloud-based platforms and autonomous equipment, further enhances operational efficiency and reduces labor costs. Additionally, hyperspectral sensing can aid in breeding programs by identifying disease-resistant soybean varieties, contributing to long-term solutions for SDS. As climate change intensifies, these technologies offer farmers the tools to adapt to shifting disease pressures, ensuring more resilient, cost-effective, and sustainable soybean production.