REAL TIME DETECTION OF FOREST FIRE USING FIRENET CNN AND EXPLAINABLE AI TECHNIQUES

Abstract

This paper presents a forest fire detection system using deep learning and explainable artificial intelligence techniques for accurate and reliable image-based classification. The proposed framework processes input images through stages including image acquisition, preprocessing, feature extraction, classification, and visualization. The system employs the EfficientNetB0 model for feature extraction and binary classification of fire and non-fire images, while Grad-CAM is used to provide visual explanations by highlighting fire-affected regions. The model is trained and evaluated using the FIRE dataset, consisting of labeled fire and non-fire images, with preprocessing techniques such as resizing, normalization, and data augmentation applied to improve performance. The extracted features are passed through fully connected layers with sigmoid activation to generate classification outputs. Experimental results demonstrate that the model achieves an accuracy of 94%, with a loss ranging between 0.04–0.085. The validation accuracy is observed between 95%, with validation loss ranging from 0.0–0.09, indicating stable learning and good generalization. The integration of explainable AI enhances transparency and trust, making the system suitable for real-world fire detection applications.