Research On Ground Condensation Image Classification Based On Deep Learning

The ground condensation weather includes dew,frost,rain and so on.Dew is an important water resource for plant growth.Frost can cause certain agricultural disasters and even plant deaths.The monitoring and identification of ground condensation weather is important for forecasting extreme weather,protecting crop yields and ecological environment.Feature extraction is difficult when traditional image processing techniques are used to identify ground condensation weather.Using deep neural network for image classification can avoid explicit feature extraction and expression,but implicitly perform feature learning in network training.It has higher fault tolerance.Firstly,this paper introduces the research background and development status of ground condensation image classification and recognition,expounds the basic ideas and training methods of deep learning,and introduces the typical convolutional neural network in deep learning.Secondly,a method of expanding deep training image set based on generative adversarial network is proposed.Based on the principle of generative networks,this method adds some convolution layers to the traditional deep-convolution generative network structure to construct a 128×128×3 dimensional output generative network model and uses it to automatically generate ground condensation images with classification information.Experimental results show that the image generated by this method is visually similar to the input original image,and has the classification feature information of the original image,which solves the problem of limited training samples.Then,a ground condensation image classification method based on multi-feature fusion of deep convolutional network is proposed.Based on the traditional convolutional network,shallow features with more detailed texture information and deep features with more explicit semantic classification information are combined by increasing multiple paths from the shallow layer to the deep layer.It is used to enhance feature information of deeper layer.The experimental results show that,compared with traditional convolutional network,this method can further improve the classification accuracy of the network.Finally,the paper is summarized and future research prospects are made.