In this paper, we propose a deep learning based smoke detection method which overcomes drawbacks of the conventional smoke detection method. Real-time smoke detection via machine-based identification method in the area of surveillance system has been of great advantage in recent era. An effective smoke detection strategy is necessary to avoid the hazard resulting from fire. The conventional smoke detection method lacks in accuracy, therefore, deep learning based smoke detection is adopted for the same purpose. However, a lot of video smoke detection approach involves minimum lighting and it can be required for the cameras to discover the existence of smoke particles in a scene. Eliminating such challenges, our proposed work introduces a novel concept like hybrid reinforcement deep Q learning classifier of smoke detection. This work takes the correct decision about the smoke particles via reinforcement deep Q learning and classifies the smoke particle with the deep convolutional neural network. The proposed real-time algorithm is aimed to provide proper education for the engineers to detect moving objects for the purpose of developing surveillance systems. This method is also helpful for the beginners who have keen interest in the field of deep learning to control fire. Here, to observe the temporal variance of fire smoke, spatial analysis is identified in the present frame and in the subsequence of the video, spatio-temporal analysis has been taken into account. Finally, smoke particles are classified with the novel reinforcement Q learning-based classifier and experimental results show a better performance regarding classification accuracy. So we can detect smoke successfully with the novel method. The main purpose of this work is to describe and formalize a machine learning -based smoke detection algorithm that can be used by students.
A Semantic Segmentation Method for Early Forest Fire Smoke Based on Concentration Weighting
