Study On Smoke Detection In Real Scene Based On Deep Learning

Smoke detection plays an important role in modern outdoor fire warning and is an important embodiment of modern fire control system.At present,more and more scholars combine deep learning to carry out smoke detection,which is faster in response and wider in detection range compared with traditional smoke detection algorithm.However,it is difficult for most algorithms to obtain both good detection rate and low false alarm rate under the complex outdoor background.Therefore,this paper conducts an in-depth research on the smoke generated in the actual scene by combining the existing smoke detection algorithm based on deep learning.The main research of this paper includes the following parts:（1）During the process of selecting the actual scene for real-time monitoring,Vibe+algorithm and cascade network structure are used to obtain the dynamic target images from the actual scene as the original data sets,which includes multi-scale and multi-form smoke images and various object images causing false positives.（2）In this paper,an improved network model based on AlexNet is proposed to extract the features of smoke images in actual scenes through the algorithm of convolutional neural network.Due to the influence of actual background and the changeful texture of smoke,the network models trained by the original public data sets cannot have good performance in actual scenes.Therefore,this paper increases the convolution layer,uses the normalized layer and overlapping maximum pool during the model and connects all the last layer to the output of two neurons.At the same time,it adjusts the parameters of network model to obtain a good effect on smoke detection.The experimental results show that the accuracy of smoke detection in this paper is improved by 3%compared with the classical network model.（3）In view of the imbalanced distribution of data sets and the lack of multiform smoke images in the actual scene,it is difficult to train a robust smoke detection model.In this paper,an improved deep convolutional generative adversarial network is adopted to expand the smoke image data sets.Compared with the original generative adversarial network,it ensures the diversity and clarity of image generation,making the data sets distribution more balanced.The experimental results show that the algorithm proposed in this paper achieves better smoke detection effect in the actual scene.