Air pollution is a major health risk and requires efficient monitoring and forecasting systems. This study uses image-based methods to overcome the limits of traditional data collection. A modified Convolutional Neural Network (CNN) classifies natural images into air quality categories based on the Air Quality Index (AQI). A comprehensive data set is assembled, comprising approximately 4000 images captured in various locations, including Tamil Nadu, Bengaluru, Mumbai, Kolkata and Durgapur, along with the corresponding measurements of PM \(_{2.5}\) , PM \(_{10}\) , CO, SO \(_{2}\) , and O \(_{3}\) . The tests show that our method achieves up to 98% accuracy, outperforming models such as random forest, support vector machine, and linear regression. This research introduces a reliable image-based approach to predicting air pollution and provides an alternative to traditional methods.