CNN based Intelligent Framework to Predict and Detect Fire

Abstract

Background/Objectives: Conventional detectors have a low degree of perfection and delicacy in relating the position of the fire and may give false admonitions. This exploration study is aimed at exercising a frame that's an agglomeration of traditional strategies and can be represented as a vision- grounded frame for intelligent finding and prediction of fire, using convolution neural systems. The performance of the said fire finding, and prediction framework is compared grounded on the perfection values under different evaluation scripts. Methods/Statistical analysis: First, image processing algorithms are used to build a fire detection framework. The continuous learning system uses the Deep CNN model and differentiates the fire data from the other data set. Better precision is gained with respect to the probability of the fire and smoke data. This helps to trigger an alarm at the right time. The Learning rate finder class will be used to derive the optimal learning rate. A continuous learning system tunes the accuracy of the classifier. When smoke is detected, movement of the smoke is taken into consideration. Using a moving object detection algorithm, suspected smoke regions are detected by the CNN algorithm. Data sets: More than twelve hundred plus augmented fire datasets were considered for this study, and two thousand plus non fire datasets were taken from the sources like ImageNet and Kaggle. Findings: Image processing technique helped to attain an accuracy of 85%. Addition of CNN improved the accuracy of the prediction to 95%. Novelty/Applications: An integrated framework enabling faster detection and prediction of fire and smoke using image processing techniques and using the processed image dataset, to classify with precision the smoke, fire, no-fire processed and augmented dataset.