Weather image classification using convolutional neural network with transfer learning

Weather condition is an important factor that is considered for various decisions. In the industrial world, weather classification is very useful, such as in development of self-driving cars, smart transportation systems, and outdoor vision systems. Manual weather classification by humans is inconsistent and takes a long time. Weather forecast information obtained from the internet is not real time at a specific location. Weather image has unique characteristics because one type of weather can be like another. Computer vision is a branch of computer science to recognize or classify images that can assist in classifying weather images that do not depend on weather forecast information from the internet. This study aims to classify weather images using Convolutional Neural Network (CNN) with Transfer Learning. Four CNN architectures, MobileNetV2, VGG16, DenseNet201, and Xception were used to perform weather image classification. Transfer learning was used to speed up the process of training models to get better performance faster. The proposed method will be applied to the weather image which consists of six classes, cloudy, rainy, shine, sunrise, snowy, and foggy were classified in this study. The experiment result with 5-cross validation and 50 epochs showed that the Xception has the best average accuracy of 90.21% with 10,962 seconds of average training time and MobileNetV2 has the fastest average training time of 2,438 seconds with 83.51% of average accuracy.