A hyperspectral-based system for identification of common bean genotypes resistant to foliar diseases in Tanzania

Abstract

Common bean is one of many legumes (family Fabaceae) widely cultivated for their edible seeds, seedpods, and leaves. Despite its benefits and life dependency especially in sub-Saharan Africa, foliar diseases are causing a loss of 20% to 80% of common bean production, and the development of improved common bean seeds resilient to those foliar diseases is still an issue where among the major problem that the bean breeders are facing is manual phenotyping; a slow field process and prone to errors as it depends on the eyes of the viewer. According to the literature, imaging technologies have been introduced to help in different processes for crops and disease management. However, there is a lack of automated mechanisms for phenotyping processes to help breeders in trait data collection, disease scare classification, and analysis of all data collected to identify resilient genotypes digitally. Among existing solutions, there is still also a gap in plant health monitoring during all its growing stages needed by breeders. Therefore, this study developed a unique hyperspectral data-based approach for identifying bean genotypes resistant to foliar diseases and plant health monitoring. Using the Random Forest classifier this study proved the genotype classification in three main breeding categories; Resistant, Medium, and Susceptible. The experiment was conducted in four Regions of Tanzania and three classifiers “Random Forest, XgBoost, and four layers Neuro Network algorithms” were trained and tested with results of 0.96, 0.95, and 0.92 respectively. The model was deployed on the cloud server where it is linked to a web application for easy classification and data analysis. Applying different vegetation indexes including the Chlorophyll Index, Photochemical Reflectance Index, Water Band Index, Modified Chlorophyll Absorption in Reflectance Index, Nitrogen Reflectance Index, Structure Insensitive Pigment Index, and Simple Ratio efficiently proved to be used for plant health insight before disease symptoms are seen. This saves breeders time, reduces errors, and helps them with digital phenotypic data, faster analysis, and easy storage for future references.