CNN-Based Detection and Classification for Real-Time Crop Monitoring

Aim: This study aims to improve the accuracy of real-time crop disease detection by comparing the traditional machine learning (ML) algorithms to Convolutional Neural Networks (CNN), along with Grad-CAM visualization for explanation. Materials and methods: The images of diseased and healthy leaves are preprocessed with contrast stretching and edge enhancement. Two groups were evaluated: Group 1 utilised 29 samples of crops in which CNN-based detection and automatically extracted features using convolutional layers were used, while Group 2 utilised the same 29 crop images in which the random forest (ML classifiers) had been trained. Models were compared in terms of accuracy, precision, recall, F1-score, and inference time, and statistical significance was set at p < 0.05 and a 95% confidence interval. Results indicated that CNN performed better than ML models with an accuracy of 92.00% compared to 84.00% and an inference time of 1.20 seconds. Grad-CAM visualization identified critical disease areas accurately, which improved model transparency. Conclusion: The research concludes that CNN is better for real-time crop monitoring with enhanced accuracy, early detection of diseases, and enhanced agricultural productivity