Classification of haze in city images with convolutional neural networks and transfer learning

Abstract

Air quality has an enormous impact on health. To take preventive measures on time, it is important to track and estimate air pollution. In the estimation of air pollution, the data acquisition from images is easy and of low-cost, when compared with sensor-based data acquisition. Machine and Deep Learning methods utilise images and videos from city cameras or social media and provide accurate estimations of air pollution. In this context, the aim of this study was testing the accuracy and efficiency of Deep Learning and Convolutional Neural Networks (CNNs) in differentiating between fog and polluted air (smog) in city images through transfer learning. The results demonstrated that Convolutional Neural Networks (CNNs) and Transfer Learning can be used as efficient methods for fog/smog classification.