Weather Classification for Multi-class Weather Image Based on CNN

Abstract

Multiple weather image classification is a very important topic in real life. Convolutional Neural Network (CNN) is a feedforward neural network that excels in image processing, but the accuracy obtained by weather image classification using simple CNN models is not very satisfactory in the previous studies. In machine learning, Support Vector Machine (SVM) is a very powerful classifier. This work proposes an effective classification method by combining CNN and SVM by taking advantage of their respective advantages. At the same time in the weather scene, the brightness of the image is also a point of concern. Hue, Saturation, Value (HSV) color model can visualize the brightness of the image, so the paper experiments on both RGB and HSV images to find which pattern of color space can achieve better result. In the experimental results, the best results are obtained by using a combination of CNN and SVM to analyze RGB images, which can achieve 77.38% in the testing dataset.