Symptom-based early detection and classification of plant diseases using AI-driven CNN+KNN Fusion Software (ACKFS)

This paper investigates and introduce an AI-driven CNN-KNN Fusion Software (ACKFS) for symptom-based early detection and classification of plant diseases. The approach integrates Convolutional Neural Networks and K-Nearest Neighbor’s to enhance classification accuracy. This research follows a structured four-phase process: pre-processing, segmentation, feature extraction, and classification. Using two datasets, ACKFS significantly improved accuracy to 94.56% and 87.52%, respectively. These results surpass the performance reported by previous researcher’s, demonstrating the effectiveness of CNN-KNN fusion for real-time plant disease detection on smart devices, contributing to precision agriculture and enhanced plant health monitoring. • Early Disease Detection to Mitigate Crop Loss: Reduces delays in mango disease detection, preventing 30% crop losses from Anthracnose. • Reduce Economic Impact: Enables rapid, automated detection for timely intervention, protecting yields and farmer livelihoods. • State-of-the-Art Accuracy: Achieves 94.56% accuracy on Mango Leaf-BD and 87.52% on Leaf Repository, surpassing MobileNet R-CNN (70.53%). • Cross-Dataset Generalizability: Performs robustly across diverse datasets, adapting to different disease complexities in mango and multi-species leaves. • Low Prediction Error: Minimizes training (0.1112), validation (0.1023), and testing (0.1059) losses, ensuring reliable real-world application. • Reduces economic impact: Enables rapid, automated detection for timely interventions, protecting yields and farmer livelihoods. • State-of-the-Art Accuracy: • Achieves 94.56% testing accuracy on Mango Leaf-BD and 87.52% on Leaf Repository datasets, outperforming prior benchmarks (e.g., MobileNet R-CNN: 70.53%). • Cross-Dataset Generalizability: • Maintains robust performance across diverse datasets (mango-specific vs. multi-species leaves), demonstrating adaptability to varying disease complexities. • Low Prediction Error: • Minimized divergence between training (0.1112), validation (0.1023), and testing (0.1059) losses, ensuring reliable deployment in real-world farming.