Research On Early Forest Fire Smoke Detection Technology Based On Video Sequence

In real life,the sudden occurrence of fire is difficult to control,especially after the occurrence of wild forest fires,not only will cause serious damage to the natural ecosystem,but also pose a serious threat to human life and property.Because smoke occurs earlier than flame,their early detection is the key to effective combat fire spread.Therefore,the development of an effective detection and early warning system for early forest fire smoke can play a vital role in protecting the ecological environment and reducing potential casualties and property losses,and has great theoretical and practical significance.This paper mainly studies the early fire smoke in the complex forest scenes in the field,which is more difficult and challenging than the indoor close-range video smoke detection.Firstly,due to the wide spatial range of the forest environment and the uncertainties of the natural environment,it is not conducive to the detection of early fire smoke in the forest.Secondly,it is difficult to detect forest fire smoke in the early field because of the long distance of smoke occurrence points,the small range of smoke occurrence,the unclear morphological characteristics and the non-rigid objects of smoke.Therefore,in the light of the technical difficulties of early fire smoke detection in complex forest scene,this paper focuses on the forest fire smoke motion region detection and feature extraction and classification recognition algorithms.The main work of this paper is as follows:(1)For the current detection methods are not flexible and recognition rate is not high,an early wildfire smoke detection method based on multi-features fusion is proposed.The method first extracts the motion foreground by an improved four-frame difference method and Gaussian Mixture Model.Then,it used linear combination of smoke color features,wavelet transform analysis and Local Binary Pattern texture features by the k-Nearest Neighbor classifier for identification.Experiments in different video scenes verify the effectiveness of the proposed method in smoke detection.(2)For the traditional methods of smoke detection and recognition are to manually extract features from the input images and then use a classifier to classify and identify the smoke,which makes it difficult to extract the effective essential characteristics of smoke,a video smoke detection algorithm based on convolutional neural network is proposed.The algorithm first narrows the scope of the smoke detection area according to the proposed moving target detection method,that is,identifies the suspected smoke area.Then,after identifying the suspicious area,the deep convolutional neural network is used to extract the feature vectors of the suspected regions,and finally the most effective features are extracted through the classification layer to achieve smoke detection.(3)Concerning the problems that convolutional neural networks can only extract the spatial features of smoke and cannot fully utilize the motion information of smoke in the video,a research method of forest fire video smoke detection based on time-space deep neural network is proposed.A video smoke detection framework using a combination of Convolutional neural networks(CNNs)and Recurrent neural networks(RNNs)is designed to detect the feature information of smoke in spatial-temporal domain.The CNNs part firstly discriminates the preliminary spatial domain features after automatically extracting features from the motion region,and then accumulates motion information between a series of consecutive frames through the motion flow networks and RNNs part to further distinguish between smoke and non-smoke regions.In this paper,compared with the existing methods using deep convolutional neural networks model,these methods only extract the spatial domain features of smoke.The experimental results show that the proposed method is higher than the CNNs model which only extracts the spatial features.Through the experiments on more than one video scene,it shows the effectiveness of the proposed method and improves the smoke detection ability.