Bitter Gourd Leaf Disease Detection Using MobileNet Architecture

Timely identification of plant diseases plays a vital role in promoting sustainable agricultural practices. In bitter gourd cultivation, early-stage diagnosis is especially critical, as it enables proactive intervention strategies to minimize potential yield degradation. This study proposes an optimized MobileNetbased deep learning model for automated leaf disease classification, achieving 94.3% accuracy, 93.2% precision, 93.5% recall, and a 93.4% F1-score with only 4.2 M parameters. Compared to conventional CNN architectures, MobileNet demonstrates superior classification performance while maintaining a lightweight structure, making it ideal for real-time applications. A confusion matrix analysis reveals that the most significant misclassifications occur in differentiating early-stage diseased leaves from healthy ones, emphasizing the need for enhanced feature extraction. The ROC analysis confirms MobileNet's robustness, achieving the highest AUC of 0.899, indicating its strong capability in distinguishing fine-grained disease variations. The integration of transfer learning from ImageNet further enhances classification accuracy from 85.6% (random initialization) to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 4. 3 \%}$</tex>, demonstrating the advantage of leveraging pretrained models for feature extraction. To enable practical deployment, the MobileNet model was integrated into an Android-based diagnostic application optimized with Tensor-Flow Lite for real-time disease detection. These results establish MobileNet as an efficient, scalable, and field-deployable solution for early plant disease detection, enhancing crop management and precision agriculture.