AN AI-DRIVEN IMAGE RECOGNITION SYSTEM FOR EARLY DETECTION OF CROP DISEASES USING A CONVOLUTIONAL NEURAL NETWORK

This study presents an AI-driven image recognition system for the early detection of plant diseases using Convolutional Neural Networks (CNNs) to mitigate crop losses and advance sustainable farming practices. Employing the publicly available New Plant Diseases Dataset from Kaggle, comprising approximately 87,000 augmented images across 38 classes (covering 14 crop species and various diseases), a custom CNN architecture was developed and trained. The model incorporates five convolutional blocks with batch normalization, ReLU activation, and max pooling, followed by a fully connected head with dropout for regularization. Trained over 15 epochs using the AdamW optimizer and cross-entropy loss on a GPU-accelerated environment, the system achieved over 98% accuracy on the validation set. Performance was evaluated through confusion matrices and accuracy metrics, demonstrating robust classification of healthy and diseased leaves. The findings highlight the potential of deep learning for timely disease identification, reducing reliance on chemical pesticides, minimizing economic losses (estimated globally at $220 billion annually due to plant diseases), and promoting eco-friendly agriculture, particularly in resource-limited settings like Nigeria. Recommendations include mobile app integration for farmer accessibility and compliance with sustainable development goals. This study has made significant strides in the development and application of artificial intelligence for crop disease detection, particularly tailored to the challenges faced in Nigerian agriculture. By leveraging a custom Convolutional Neural Network (CNN) architecture trained on the New Plant Diseases Dataset, which encompasses over 87,000 images across 38 classes of healthy and diseased plant leaves from 14 crop species, the model achieved an exceptional accuracy of 98.5% in multiclass classification tasks. This performance surpasses established benchmarks in similar studies, such as the 97.3% accuracy reported in automated plant disease detection using CNNs on comparable datasets, and the 95% accuracy in AI-driven image annotation for plant diseases via Google Cloud Vision.