WeatherNet: Transfer Learning-based Weather Recognition Model

Abstract

A transfer learning (TL) based multi-classification model has been developed to classify and recognize collected weather dataset with 1000 different weather images belonging to four different classes of weather. MobileNet V2 has been applied as a pre-trained model with a combination of weather image classifiers which results in the best recognition accuracy of 98.25% in the case of rainy (R) class images. Methodological techniques and challenges encountered while experimenting has also been presented in the detailed description. Along with this, the proposed model has also been compared with a simple convolutional neural network (CNN) model which results in outperformance of the TL model in terms of efficiency and efficacy.