A Comprehensive Machine Learning Approach for Wild Fire Detection and Classification Using Squeezenet and Ridge Regularized Logistic Regression

Wild fire is a serious environmental and socioeconomic menace and therefore any effort aimed at the early and accurate detection of wild fire has to be very useful. This paper proposes a novel machine learning framework for wildfire detection as well as its classification utilizing SqueezeNet as a lightweight convolutional neural network (CNN) and ridgeregularized logistic regression (RR-LR) as a classifier to improve the classification performance and model stability. The proposed method is compared with a set of machine learning classifiers like kNN. Al deliberation of these experimentation results suggests that the ridge-regularized logistic regression model performs better than these methods, specifically in cases concerning overfitting and in situations where the amount or balance of data available is small. Furthermore, SqueezeNet achieves efficient feature extraction that directly lessens the computation cost so users can deploy it in real-time especially in areas with limited resources. This work reveals the benefits of combining deep feature extraction with proper regularization methods and points to a promising solution for the wildfire detection and classification problem.