Deep Learning Algorithm for Fire Detection

Based on data from ayobandung.com, the Bandung City Fire and Disaster Management Office (Diskar PB) handled 199 fires in 2019. Of these, as many as 121 cases of building fires, 69 bush or grass fires, and the rest are handling fires outside the city of Bandung. As a result of these fires, it is estimated to result in losses of IDR 44,363,700,000. However, with the blackout carried out by officers, Diskar PB managed to save material worth IDR 810,790,534,400. According to them, the obvious dominant cause is human negligence, starting from the stove and then electricity. The speed with which people are aware of fires and reporting fire incidents also affects. The time for making this report is precious for Diskar PB officers. So it is crucial to get information as soon as possible before the fire spreads more. For indoor fire detection, a system is designed with sensors and cameras. This system will detect in real-time if there is smoke or fire, the system will notify the user by telegram. This fire system uses the Backpropagation and Convolutional Neural Networks (CNN) method used to carry out object recognition and fire patterns. This system can improve safety in fire prevention. The fire detection system created has an accuracy rate of 95% for the Backpropagation method. Meanwhile, CNN has an accuracy rate of 97%.