A robust framework combining hyperspectral imaging and machine learning for assessing sudden death syndrome (SDS) severity in soybean foliage

Authors: IQBAL, ZAFA - University Of Florida; BURKS, THOMAS - University Of Florida; YADAV, PAPPU - South Dakota State University; OBELLANENI, SATYA - University Of Florida; RASOOL, INAYAT - South Dakota State University; FREDERICK, QUENTIN - University Of Florida; Qin, Jianwei; Kim, Moon

Submitted to: Journal of Biosystems Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/1/2025
Publication Date: N/A
Citation: N/A

Interpretive Summary: Sudden Death Syndrome (SDS) is a serious disease that affects soybean plants and can cause significant losses for farmers. Currently there’s no good way to stop it with chemicals or natural treatments. Accurate detection and severity assessment can help mitigate losses through preventive measures and development of resistant varieties. However, figuring out the gravity of the disease usually takes a lot of time and requires experts. This study developed a new method to detect SDS and measure how severe it is using advanced imaging. Images at a series of different wavelengths were collected from both sides of 284 soybean leaves from both healthy and SDS-affected. We developed AI computer programs to differentiate healthy and SDS-affected leaves with an accuracy of 97%. To evaluate severity of SDS, we further classified regions of abnormal leaves into four categories (unaffected, mildly affected, severely affected, and dehydrated) with an accuracy of 84%. This advanced imaging and AI technology can help farmers accurately assess SDS severity for better crop management and support breeding programs in developing SDS-resistant varieties to enhance productivity and disease control.

Technical Abstract: Sudden Death Syndrome (SDS) in soybeans causes significant yield losses, with no effective biological or chemical control. Accurate detection and severity assessment can help mitigate losses through preventive measures and the development of resistant varieties. Traditional severity assessment is labor-intensive and requires expertise, highlighting the need for automated and precise detection. This study aimed to develop an automated workflow to detect SDS and classify severity in soybean leaves. Hyperspectral reflectance images (398-1011 nm) were captured for 284 leaves (both healthy and SDS-affected) from both sides. Five distinct datasets were created using key spectral bands, and a pseudo-RGB dataset for model evaluation. A YOLO11 model was trained and fine-tuned to distinguish healthy leaves from diseased ones, achieving a mean average precision (mAP) of 97.10%. To further classify regions of abnormal leaves into categories (unaffected, mildly affected, severely affected, and dehydrated), both supervised and unsupervised methods were employed using reflectance and NDVI values. Among these methods, unsupervised k-means clustering using reflectance data demonstrated superior performance, achieving an overall accuracy of 84.24%, with 13% of dehydrated pixels misclassified as severe, 17% of mild pixels as unaffected, and 29% of unaffected pixels as mildly affected. Future work will seek to evaluate the progression of SDS throughout the growing season, which was a limitation of this study due to the late season data collection. The findings from this study could help farmers accurately assess SDS severity for better crop management and support breeding programs in developing SDS-resistant varieties, enhancing productivity and disease control.