Research On Fire Image Classification And Detection Algorithm Based On Deep Learning

With the development of the times,a variety of fire detection equipment and methods have emerged.In order to improve the accuracy of fire identification and reduce the rate of missing reports,image-based fire detection method is becoming more and more popular.The development of deep learning also provides new opportunities and challenges for fire recognition.This paper mainly studies the application of classic convolutional neural network classification model and target detection model to fire data.According to the characteristics of fire data,we use suitable methods to achieve the accurate classification and detection of fire image data.Aiming at the fire classification and recognition network,the fire,smoke and other textured data are analyzed,and the classic alexnet is improved.In order to learn the distinguishing features,a rotation invariant convolutional neural network is designed,which can be applied to the fire image accurately.The classification of the fire image is distinguished,and the accuracy is 98.8%,compared with the ordinary Alexnet has increased by 3.2%.For the fire detection and recognition network,the classic Faster R-CNN is improved.For the feature extraction network,Resnet101 is used as the feature extraction network,and FPN is used to extract the shallow and high-level features of Resnet101,then entering the perception module to extract a variety of convolution features,and using pixel attention mechanism and channel attention mechanism to strengthen the target position and weaken the rest of the part,which improves the detection accuracy of the fire data set.Finally,the average detection accuracy map of the model is 85.1,which is 9% higher than the Resnet101 feature extraction network,and 5% higher than the Resnet101 feature pyramid network.