A Data Fabric Approach to Crop Yield Optimization Through Efficient Cross Species Diagnostics

Crop diseases pose a challenge for global food security, leading to annual yield losses which lie between 20% and 40%. This issue is even more concerning for developing countries like Liberia, Somalia and Bangladesh. This is because agriculture serves as a cornerstone of both the national economy and individual livelihoods in these countries. In recent years, deep learning has shown promising results in automating crop disease detection. However, these advances are largely limited to major industrial crops with extensive annotated datasets. Neglected and Underutilized Species (NUS) often lack large, well-labelled datasets, making reliable diagnostic systems difficult to develop. This research introduces a novel framework that combines data fabric architecture with transfer learning to develop accurate disease detection systems even when training data is scarce. The proposed data fabric achieved an improvement of +0.3801 in F1-score for Cassava. Other notable improvements were observed for Radish (+0.2686) and Pigeonpea (+0.2133). Identification of the best knowledge donor through cosine similarity was also performed, with Holy Basil showing the highest knowledge transfer gain (+0.2800).Additionally, the framework demonstrated the ability to classify disease severity and identify the closest matching disease from the relational fabric for previously unseen conditions. The study also investigates the causes of negative knowledge transfer within cross-species diagnostic systems.