Early Detection of Plant Stress Using GLCM Feature Extraction and Hybrid Random Forest-Naive Bayes Classifier

Aim: The project seeks to create an automated image processing system for detecting leaf diseases and environmental stress in plants. It offers real-time monitoring to help improve seedling growth and environmental conditions. The system prioritizes speed, cost-effectiveness, and high accuracy for efficient crop management. Materials and methods: This study includes two groups: Group 1 employed a GLCM-based classification model, whereas Group 2 used a Naive Bayes model, both with 26 samples. Color, texture, and morphological features were retrieved from photographs captured daily using an RGB camera over a two-week period. K-means segmentation, noise reduction, and feature extraction were the preprocessing steps for model training. Results: The GLCM model (82% – 86%) performed worse than the Naive Bayes model (95%), which was used to detect leaf diseases. The model's durability was demonstrated by cross-validation, which produced MAE scores of 0.138 and 0.163. The system provides a low-cost, automated method of detecting diseases early on, hence enhancing crop yield. Conclusion: In this work, it is observed that the Naive Bayes model has significantly higher accuracy in detecting leaf diseases compared to the GLCM model.