An image based case based reasoning system to identify maize crop attacks and recommend treatments

Abstract

Maize (Zea mays) is a critical global staple, essential for food, livestock feed, and industrial applications. Its economic significance requires effective management to ensure high yields and quality. However, maize is highly susceptible to various pests and diseases that significantly impact productivity. Traditionally, farmers rely on human experts for crop protection, but access to such expertise is limited. This study leverages the Case-Based Reasoning (CBR) paradigm, which relies on knowledge from previously solved problems to address current ones, to profile maize disease attacks. The designed and developed CBR system diagnoses Common Rust (CR), Gray Leaf Spot (GLS), and Blight (Bl) based on leaf images, extracts features based on the VGG16 model and uses machine learning techniques in the different layers to recommend treatment actions and preventive measures. An experiment has been conducted on 4188 images that included 1306 CR, 574 GLS, 1146 Bl, and 1162 healthy leaves. The system achieved a balanced accuracy of $$92.49\%$$ in disease detection under the split of $$80\%$$ of training and $$20\%$$ of testing samples, outperforming similar systems. This enhanced performance is due to the effective integration of historical cases, represented by global features of images of previously attacked maize leaves.