Multi-Class Weather Classification Using ResNet-18 CNN for Autonomous IoT and CPS Applications
Qasem Abu Al‐Haija, Mahmoud A. Smadi, Saleh Zein-Sabatto
Abstract
Severe circumstances of outdoor weather might have a significant influence on the road traffic. However, the early weather condition warning and detection can provide a significant chance for correct control and survival. Therefore, the auto-recognition models of weather situations with high level of confidence are essentially needed for several autonomous IoT systems, self-driving vehicles and transport control systems. In this work, we propose an accurate and precise self-reliant framework for weather recognition using ResNet-18 convolutional neural network to provide multi-class weather classification. The proposed model employs transfer learning technique of the powerful ResNet-18 CNN pretrained on ImageNet to train and classify weather recognition images dataset into four classes including: sunrise, shine, rain, and cloudy. The simulation results showed that our proposed model achieves remarkable classification accuracy of 98.22% outperforming other compared models trained on the same dataset.