Image-Based PM2.5 Estimation From Imbalanced Data Distribution Using Prior-Enhanced Neural Networks

The effective monitoring of PM2.5, a major indicator of air pollution, is crucial to human activities. Compared to traditional physiochemical techniques, image-based methods train PM2.5 estimators by using datasets containing pairs of images and PM2.5 values, which are efficient, economical, and convenient to deploy. However, existing methods either employ handcrafted features, which can be easily influenced by the image content, or require additional weather information acquired probably by laborious processes. To estimate the PM2.5 concentration from a single image without requiring additional data, we herein propose a prior-enhanced (PE) framework that learns from both the input image and the corresponding prior maps of dark channel (DC) and inverted saturation (IS). We implemented three types of PE networks by incorporating the most recent baselines, demonstrating the versatility of the proposed framework for different network models. In addition, we propose a histogram smoothing (HS) algorithm to address the issue of imbalanced data distribution, thereby improving the estimation accuracy in cases of heavy air pollution. To the best of our knowledge, this study is the first to address the phenomenon of a data imbalance in image-based PM2.5 estimation. Finally, we construct a new dataset containing multiangle images and more than 30 types of air data. Extensive experiments across diverse datasets demonstrate that our method excels in accurately estimating PM2.5 concentrations from images in an end-to-end manner, outperforming state-of-the-art image-based approaches. The new dataset and codes are available at https://github.com/lizhangray/PE-PM2.5.