Fire Detection and Prediction Framework Using Vision Based System and Convolutional Neural Network
Conventional sensors have a relatively low degree of precision and accuracy in identifying the location of fire and may give false alarms. This research study is aimed at utilizing a framework that is an agglomeration of traditional strategy, a vision-based framework for intelligent detection and prediction of fire using convolution neural systems. The performance of the Fire Detection and Prediction Framework is compared based on the precision values under different evaluation scenarios. This research is two-phased - the first phase is used to build a fire detection framework or tool based on image processing methods. In the subsequent steps, the Deep CNN architecture will trigger an alert in case of fire data and will improve on the classification by continuous learning of fire and no fire data. The Learning Rate Finder class will be used to find out the optimal learning rate, and the accuracy of the classifier is then subsequently defined. More than 1200 plus augmented fire datasets were considered under study, and the non-fire dataset constitutes 2000 plus entries from ImageNet and Kaggle. The problem has two classes of data, i.e., fire and non-fire and smoke. The learning accuracy achieved is 90-92%, detection via a vision-based approach attained is 85%, and fire and smoke prediction attained using Convolutional Neural Network is 95%. The error rate was 0-0.5%.