A Transfer Learning Method for Meteorological Visibility Estimation Based on Feature Fusion Method

Meteorological visibility is an important meteorological observation indicator to measure the weather transparency which for transport safety. It a challenging problem estimate visibilities accurately from image characteristics. This paper proposes transfer learning method estimation based on feature fusion. Different existing methods, proposed estimates data processing and features’ extraction in selected subregions of whole therefore it had less computation load higher efficiency. All database images were gray-averaged firstly selection effective features extraction. Effective are extracted static landmark objects can provide useful information estimation. Four different methods (Densest, ResNet50, Vgg16, Vgg19) used subregions. The by neural network then imported into support vector regression (SVR) model, derives estimated Finally, weight fusion subregion models, overall comprehensive was image. Experimental results show that accuracy more than 90%. image, with high robustness effectiveness.