Efficient Weather Detection Using Lightweight Convolutional Neural Networks suitable for Implementing and Deploying on Edge Devices

Weather detection is vital in many fields, including disaster recovery, farming, transportation, and climate studies. Making educated decisions and reducing potential dangers requires the quick and precise observation of weather patterns. As CNN-based deep learning models automatically train and extract significant features from raw image data, they often achieve higher accuracy in a variety of image classification tasks and exhibit good data generalization. In addition to the accurate detection, it should be lightweight to deploy to edge locations. To address this, we have developed a Smart lightweight deep-learning model to detect the weather accurately and it can be conveniently deployable in resource-constrained devices. This study uses several deep lightweight learning models, including MobileNetV2, MobileNetV3_Small, and NasNetMobile, where MobileNetV3_Small is the most efficient regarding weather detection and computational complexity reduction. The proposed approach utilized a Multi-class Weather Dataset for Image Classification for building Smart lightweight deep-learning models. Besides the weather classification, the proposed approach also gives the confidence score for the predicted climate, which provides a numerical indicator to evaluate how likely the predicted climate is to occur.