Smoke Image Detection Based On Deep Transfer Learning

The burning of straw in rural areas will cause a lot of smoke to cause haze and even cause fire.Traditional fire detection methods are not suitable for monitoring large-scale scenes in rural areas.This paper proposes a smoke image detection method based on deep learning transfer learning.As one of the obvious features of fire early,smoke detection and detection can effectively warn fires and reduce the loss of life and property.With the improvement of computer computing power and the arrival of the era of big data,deep learning has received extensive attention,and it has shown good results in various intelligent fields.The deep learning method is different from the traditional framework.It does not need to manually design the detection operator.It can automatically learn the characteristics from the data and is continuously optimized through training.Neural network training requires a lot of data support,but in some cases data collection is very difficult,such as collecting rural smoke videos and pictures.At present,there is no large-scale smoke video library on the Internet.The amount of data does not support the training of deep networks from scratch.Therefore,the transfer learning method is adopted to solve the problem of small smoke data sets.Transfer learning is a machine learning technique that applies the source domain model to the target domain.In the field of deep learning image processing,convolutional neural networks usually detect the edge,shape,and then the specific features of the target.In fact,the image characteristics of the early layer detection of the neural network are not much different,which is very suitable for transfer learning technology.Therefore,transfer learning is particularly effective in the field of computer vision.The method uses the TensorFlow framework to reference the Inception-v3 network model that has been trained on the ImageNet dataset as the source model,and uses the edges,colors,textures and other features extracted in the ImageNet dataset to construct a new smoke detection model.The verification image of the rural scene is used for verification test,and the model is analyzed and adjusted according to the result.The results show that the model detection accuracy is greatly improved compared to the traditional method.