| The invention and use of fire has brought great convenience to people's production and life,and is one of the important signs that human society has entered the era of civilization from then on.If used improperly,it will also bring hidden dangers to people's lives and property.Because the sensor has the advantages of high cost performance and easy installation,the early fire detection and alarm are mainly realized by the sensor method.However,the sensor itself has many shortcomings such as slow response time,low sensitivity,and difficulty in installation in a relatively wide space,and it is difficult to meet the needs of early detection and early warning of fire occurrence.Under normal circumstances,smoke will be generated in the early stage of the fire.If the fire danger can be found early through the analysis of the smoke characteristics before the fire,the threat to people's lives and property safety can be greatly reduced.In view of the fact that the distance of the field smoke is far away and the background cloud fog causes the existing algorithms to have low accuracy and high false detection rates,this paper systematically studies and studies the field smoke recognition algorithm and system design based on depth characteristics The main work is summarized as follows:(1)In view of the relatively small number of publicly available smoke data sets for forests,and most smoke data sets only include dense smoke data sets,this paper establishes a smoke recognition data set containing light smoke and dense smoke.In view of the shortcomings of the existing smoke data set,the original data set for smoke identification was obtained through various smoke data collection methods such as crawling the smoke data set on the Internet,shooting the smoke video in the field,and calling the scenic spot monitoring video.The smoke is subdivided into thick smoke and light smoke.Secondly,for the problem that the original smoke data set is few,the original data set is appropriately expanded.The smoke identification data set finally established can not only fully meet the research needs of the subject,but also has the research characteristics of this article.(2)In order to meet the needs of field-oriented smoke detection,this paper designs an improved VGG16C-smoke method based on cavity convolution.In this paper,the HOG feature and the LBP feature are first fused for field smoke recognition.The recognition accuracy has been improved to a certain extent.However,the light smoke recognition accuracy is low.Secondly,this paper focuses on the field smoke recognition algorithm based on deep feature extraction.Deep features can avoid the subjectivity of artificial feature extraction and can extract the most suitable features for smoke.Experimental tests show that,compared with manual feature extraction and recognition of smoke,VGG deep features can better identify smoke,especially light smoke.However,the light smoke is not obvious in the image at the beginning of the fire,the receptive field of the VGG16 network is limited,and the hollow convolution can increase the range of the receptive field without increasing the pooling layer.At this time,the light smoke can be displayed better.In this paper,a VGG16C-smoke method is designed by adding void convolution to the VGG16 network.Experimental tests show that the VGG16C-smoke network proposed in this paper can achieve 94.8%accuracy for light smoke.(3)Based on the above work,a field smoke recognition system based on depth features is implemented.The wild smoke recognition system based on depth features designed in this paper mainly includes three modules:smoke data preprocessing,smoke detection and smoke recognition and alarm.The system design includes extraction of suspected smoke areas,preliminary screening of smoke based on the HSV channel,and the type of smoke is subdivided into dense smoke and light smoke.Finally,the system function and performance of the designed system are tested comprehensively,and the system can better identify light smoke and dense smoke. |
PDF Full Text Request