A Lightweight and Explainable DenseNet-121 Framework for Grape Leaf Disease Classification
A lightweight AI just beat major models like ResNet and VGG for agricultural disease detection.
Deep Dive
Researchers have developed a new, optimized DenseNet-121 AI framework that can classify grape leaf diseases with 99.27% accuracy and an F1 score of 99.28%. It significantly outperforms established models like ResNet18 and VGG16, achieving an inference time of just 9 seconds. The model uses Grad-CAM for explainability, highlighting disease-relevant features like lesions, and is designed to be computationally inexpensive for real-world, scalable deployment in vineyards.
Why It Matters
This provides farmers with a fast, accurate, and interpretable tool for early disease detection, potentially saving crops and boosting agricultural yields.