Fire Detection Using Logistic Regression with GLCM, RGB Ratio, RGB Intersection, and Color Moments

Abstract

Fires pose a significant threat to human safety and property, particularly in densely populated urban environments where rapid and accurate early detection is critical. This study proposes an automated fire detection system based on computer vision and Logistic Regression classification, utilizing a combination of texture and color-based features to improve detection performance. The proposed approach integrates Gray-Level Co-occurrence Matrix (GLCM), RGB Ratio, RGB Intersection, and Color Moments to extract discriminative features from fire and non-fire images. The dataset, obtained from Kaggle, was preprocessed through HSV-based color segmentation to isolate candidate fire regions before manual annotation. The extracted features were then used to train a Logistic Regression model with hyperparameter tuning of the max_iter parameter to achieve optimal convergence. Experimental results show that the proposed model achieved an accuracy of 86% and a recall of 84% on the training dataset, and an accuracy of 87% with a recall of 82% on the test dataset. Despite these promising results, some false negatives were observed, indicating the need for further refinement to improve sensitivity. Comparative evaluation with a Convolutional Neural Network (CNN) demonstrated that the Logistic Regression approach achieved higher average processing speed, reaching up to 16.2 FPS for video input, compared to 11 FPS for CNN, making it more suitable for real-time applications. Overall, the integration of multi-feature extraction with Logistic Regression offers a balance between accuracy and computational efficiency for early fire detection in real-world scenarios.