Straw Burning Smoke Detection Algorithm Based On Twin Network And Feature Map Fusion

Straw burning is a common straw treatment method.The burning of straw will pollute the environment and cause traffic accidents.The hydrocarbons produced by the burning of straw will also endanger human health.Detecting and stopping burning behavior in time is of great significance for protecting human health,reducing environmental pollution and preventing traffic accidents.It is an important means to realize the detection of straw burning by identifying the smoke produced during the burning of straw.The scenes where straw burning occurs are different,and the interference of objective factors on smoke detection is different in the scenes of houses,farmland,and lakes.In addition to the interference of scene factors,features such as smoke color and gradient are not obvious,which will also bring difficulties to smoke detection.Traditional smoke detection algorithms are difficult to effectively extract the features of smoke,and the accuracy of the existing convolutional neural network methods in smoke detection needs to be improved.This paper studies the straw burning smoke detection algorithm based on the fusion of twin network and feature map.In order to reduce the impact of excessive scene differences on the smoke detection efficiency,the smoke detection scenes are divided into three categories:houses,farmland,and lakes.For each scene,a smoke detection algorithm based on multi-scale feature map fusion is used to establish a smoke detection model.When detecting,the lightweight ResNet twin network is used to match the scene of the image to be detected.According to the matching result,the smoke detection model in the corresponding scene is used to detect smoke in the image.The main work is as follows:(1)Establish a dataset of straw burning scenes.This paper collects a large number of straw burning sample sets from the cameras deployed on the signal tower,and divides these samples into three types: houses,farmland,and lake scenes according to the scenes of straw burning.(2)A multi-scene matching algorithm based on the lightweight ResNet twin network is proposed.The background factors in the scene will interfere with the straw burning smoke detection.The lightweight ResNet twin network is used to match the scene of the image to be detected,which not only effectively reduces the scene interference,but also speeds up the detection speed while maintaining the detection accuracy.Experimental results show that the accuracy of the multi-scene matching algorithm proposed in this paper is 94.53%.(3)A smoke detection algorithm based on multi-scale feature map fusion is proposed.Existing smoke detection methods are difficult to meet the demand for detection accuracy in actual applications.This paper uses the bottom-up feature extraction path of the convolutional layer to fuse the feature maps obtained by each convolutional layer,and the fused feature maps Used for the training of the detection model to improve the accuracy of the detection model.Practical application shows that the method proposed in this paper achieves an accuracy of90.56%,which is higher than the detection accuracy of the existing convolutional neural network.