Study On Classification Of Cloud Background Images By Machine Learning

Cloud plays an important role in climate prediction and meteorological services.The existing space-based or ground-based cloud image classification tasks are mostly artificial,consuming a lot of manpower,and the classification results are ambiguous because of the different standards of the observer.Under this background,this paper takes two cloud image datasets TCdata and SCdata as data sources,and studies the automatic classification method of cloud background image based on machine learning.The research work is carried out in the following four aspects.(1)In the study of the image classification method based on convolution neural network,the structure and training process of the image classification method based on the convolution neural network are studied,and the training platform of the convolution neural network is built in the Tensor Flow development environment.(2)In the study of the production and processing of datasets,two cloud image datasets of space-based and ground-based datasets TCdata and SCdata are made,and data preprocessing,data enhancement processing,dataset partition and dataset storage are processed for the two datasets.These two datasets consist of feature complexity of cloud images,providing data support for comparative experimental research of convolution neural networks with other different structures.(3)In the study of the design and improvement of the simple image classification model,the STCnet model is designed based on Le Net and the Alex Net,and the classification accuracy on the TCdata and SCdata datasets can reach more than 96%.Experiments show that the pooling layer plays an important role in the STCnet model,and the pooling layer is preserved after all convolution layers in the model.The activation function of STCnet model is improved,which improves the accuracy of cloud image classification.(4)In the study of the design and improvement of the complex image classification model,an improved Goog Le Net model for cloud image classification is designed.By combining the convolution neural network model used for feature extraction and the SVM classifier,the Goog Le Net-SVM model is designed.By comparing the classification effects of each model,it is proved that the Goog Le Net-SVM model can not only obtain superior classification accuracy,but also have the advantage of employing less consumption in classification processing.The machine learning classification method studied in this paper can be applied to the field of data processing and classification of cloud background images.It provides a theoretical support for the optimization of the classification model.