A Comparative Assessment of CNN-Sigmoid and CNN-SVM model for Forest Fire Detection

Forest fires present a notable danger to the environment, wildlife, and human lives. Early detection of forest fire is crucial to prevent their spread and minimize damage. The timely detection and response to forest fires are paramount to mitigating their destructive impact. Yet, conventional methods like visual inspection and ground patrols are labor-intensive, time-consuming, and demand extensive coverage. Recently, machine learning algorithms for image detection have emerged as an effective approach for detecting forest fire using forest fire images taken from drone. This study aims to evaluate the effectiveness of Convolution Neural Network with Sigmoid (CNN-Sigmoid) and Convolution Neural Network with Support Vector Machine (CNN-SVM) model to detect forest fire (on Forest Fire Dataset). Through our experimental result we have found that the extraction using CNN and classification using SVM has improved the training accuracy (99.87%) and validation accuracy (97.11%) of CNN-SVM model compared to CNN-Sigmoid. This observation was substantiated by the CNN-SVM achieving a test accuracy of 96.84%, while the CNN-Sigmoid attained a test accuracy of 95.79%.