Efficient Detection of Crop Leaf Diseases: A Lightweight Convolutional Neural Network Approach for Enhanced Agricultural Productivity

Crop leaf diseases pose a significant and persistent threat to agricultural productivity and food security, particularly in Bangladesh, where agriculture plays a crucial role in the economy. Effective methodologies for early disease detection and mitigation are vital in this context. However, existing literature in this field predominantly relied on computationally demanding transfer learning (TL) techniques. A lightweight custom convolutional neural network (CCNN) model is proposed for crop leaf disease detection, addressing this challenge directly. The proposed approach is evaluated using three crop leaf detection datasets, and the results demonstrate that the model achieved impressive accuracy while utilizing fewer parameters. A comparative analysis has been conducted with various transfer learning (TL) models to comprehensively compare accuracy, parameter count, and training time. Gradient Activation Map (Grad-CAM) techniques have also been utilized to generate heatmaps highlighting the focus regions during detections. The findings of this research enhance crop leaf disease detection methods, offering practical applications for improving agriculture and reducing crop disease impact.