Cost-effective and scalable urban air quality monitoring using image-based deep learning.

Air pollution is a persisting problem for the environment, mainly in urban areas, given heavy industrialization and vehicular emissions and urbanization. Exposure to harmful air pollutants, including PM2.5, PM10, and toxic gases, is a major contributor to respiratory and cardiovascular diseases, accounting for millions of deaths globally each year. Answer monitoring is, therefore, the first effective measure in limiting the health effects of air pollution. The idea of this study is to advance an effective, scalable image-based deep learning (DL) system to detect air pollution levels within urban areas. The model integrates CBAM with InceptionV3 to enhance feature extraction, improving accuracy and robustness in detecting varying levels of air pollution from images. The proposed work uses Air Pollution Image Dataset from Kaggle, containing urban pollution images from India and Nepal. Preprocessing includes Gaussian Filtering to reduce noise, Histogram Equalization (HE) to enhance contrast, Thresholding to segment image, and resizing to standard dimensions. The model employs InceptionV3 integrated with Convolutional Block Attention Module (CBAM) for enhanced feature extraction. This approach classifies images into three pollution levels: Moderate, Unhealthy for Sensitive Groups (USG), and Unhealthy, enabling efficient air quality monitoring. The CBAM-InceptionV3 model achieved strong performance, with precision, recall, and F1-scores exceeding 0.95 for each class. The proposed CBAM-InceptionV3 model achieves an overall classification accuracy of 95.23%, with precision, recall, and F1-score values consistently above 95%, demonstrating its effectiveness for air pollution level classification.