Deep learning–driven hyperspectral imaging for drought stress detection in dragoon lettuce for space production

Authors: KIM, HANGI - Chungnam National University; PARK, EUN-SUNG - Chungnam National University; Kim, Moon; Baek, Insuck; CONSTINE, BLAKE - Kennedy Space Center; SPENCER, LASHELLE - Kennedy Space Center; O'ROURKE, AUBRIE - Kennedy Space Center; LEE, HOONSOO - Chungbuk National University; KIM, GEONWOO - Gyeongsang National University; MO, CHANGYEUN - Kangwon National University; HO, BYOUNG-KWAN - Chungnam National University

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/6/2025
Publication Date: 12/18/2025
Citation: Kim, H., Park, E., Kim, M.S., Baek, I., Constine, B., Spencer, L., O'Rourke, A., Lee, H., Kim, G., Mo, C., Ho, B. 2025. Deep learning–driven hyperspectral imaging for drought stress detection in dragoon lettuce for space production. Computers and Electronics in Agriculture. 242. Article 111321. https://doi.org/10.1016/j.compag.2025.111321.
DOI: https://doi.org/10.1016/j.compag.2025.111321

Interpretive Summary: As space missions become longer, astronauts will need to grow their own food in space. To do this successfully, it is essential to keep plants healthy and quickly identify any signs of stress, such as drought. Early detection enables timely action to protect plant growth and ensure a steady food supply. This study tested an advanced camera system called hyperspectral imaging. Unlike regular cameras, it captures detailed light patterns that reveal early signs of plant stress. To improve detection accuracy, researchers integrated artificial intelligence (AI) into the analysis process. We used deep learning models known as Vision Transformers. These models were trained to recognize specific patterns in the light data and accurately determine whether a plant was healthy or experiencing drought stress. The model performed exceptionally well, achieving an accuracy of 0.954, a precision of 0.966, and a recall of 0.941. These results demonstrate that the model was highly effective in detecting plant stress, with minimal false positives or negatives. Additionally, a technique called Integrated Gradients was used to highlight specific areas of the plant affected by stress. This allowed researchers to visualize how stress developed at the leaf surfaces. Although this system was designed for space farming, it also has important applications on Earth. The technology can be used to help farmers monitor crops more precisely and respond more promptly to drought or disease conditions.

Technical Abstract: Sustainable plant cultivation is critical for supporting long-duration space missions by ensuring reliable food production in extraterrestrial environments. Early and accurate detection of plant stress is essential for optimizing growth conditions, maintaining plant health, and maximizing productivity in controlled cultivation systems. This study employed a next-generation hyperspectral imaging (HSI) system developed for space-based plant monitoring. Integrating a compact hyperspectral camera with broadband LED lighting in an overhead imaging setup enables the acquisition of reflectance spectra data, allowing for precise detection of drought-induced stress markers while minimizing heat generation and structural constraints. An AI-driven hyperspectral analysis framework was implemented to enhance automated stress detection, leveraging Vision Transformer (ViT) and Spectral Attention Module-enhanced ViT (SAM-ViT). Experimental results demonstrate that deep learning models outperform traditional machine learning approaches in hyperspectral drought stress classification. The SAM-ViT-3PE model achieved high accuracy (0.954), precision (0.966), and recall (0.941), surpassing conventional analysis methods and ensuring robust stress detection with minimal false positives and negatives. Additionally, a novel Integrated Gradients (IG) approach was employed to visualize plant stress, offering insights into stress progression at the tissue level. These results demonstrate the potential of AI-enhanced hyperspectral imaging for autonomous and precise detection of plant stress, thereby advancing sustainable space farming and supporting future extraterrestrial missions.