Deep WeatherNet: A Scalable Deep Learning Approach for Weather Image Classification on Large Datasets
Abstract
Classifying weather conditions from outdoor images can have various applications such as improving road safety, scheduling outdoor activities, and enhancing the reliability of vehicle assistant driving and outdoor video surveillance systems. However, traditional methods of weather classification involved the use of expensive sensors and extensive manpower, which were both time-consuming and tedious. Automating the task of weather classification from images can save time and resources. In this paper, we propose a framework for classifying weather images using transfer learning techniques. Our approach involves using pre-trained deep CNN models to learn features, thereby reducing the time required for classification. We also recognize that the size and quality of the training dataset are critical factors in the efficiency of the model. Therefore, we implemented our framework using the Spark platform, making it scalable for big datasets. We conducted extensive experiments on a weather image dataset, and our results demonstrate the reliability of our proposed framework. We concluded that weather classification using the InceptionV3 model and Logistic Regression classifier yields the best results, with a maximum accuracy of 97.77%. Our framework has potential applications in various fields such as agriculture, aquaculture, transportation, tourism, etc.