Cucumber Leaf Disease Detection and Classification Using Artificial Intelligence

Abstract

Cucumber (Cucumis sativus L.) is a vital vegetable crop in Bangladesh, highly valued for its nutritional benefits and economic significance. However, cucumber cultivation is severely threatened by various leaf pathogens such as powdery mildew, downy mildew, anthracnose, and bacterial wilt; which drastically reduce crop yield and quality. Early and precise disease identification is critical for implementing targeted interventions and minimizing agricultural losses. This thesis presents an advanced, Artificial Intelligence (AI)-based diagnostic system for the automated detection and classification of cucumber leaf diseases. The proposed framework evaluates a custom Convolutional Neural Network (CNN), standard transfer learning architectures (MobileNetV2, ResNet50, and EfficientNetB0), and a hybrid deep learning model combining EfficientNet with a Support Vector Machine (EfficientNet SVM). To train these models, a comprehensive dataset of labeled leaf images was compiled from both online repositories and real-field environments in Bangladesh, and subsequently optimized using resizing, normalization, and data augmentation techniques. The models were rigorously evaluated using standard performance metrics, including accuracy, precision, and recall. Experimental results demonstrate that the proposed hybrid EfficientNet-SVM model achieved the highest overall performance, yielding an exceptional 99.09% accuracy, 99.00% precision, and 99.00% recall. Among the end-to-end deep learning architectures, MobileNetV2 and the Custom CNN performed robustly (achieving 96.17% and 95.96% accuracy, respectively), while traditional ensemble methods like Random Forest (19.33%) and deeper architectures like ResNet50 (60.57%) struggled to generalize. The superior performance of the hybrid model, alongside the lightweight footprint of MobileNetV2, highlights the immediate viability of deploying these systems via mobile applications. This research offers a transformative tool for Bangladeshi farmers to rapidly diagnose crop diseases, reduce chemical pesticide overuse through precise targeting, and foster sustainable, technology-driven agricultural practices.