Air quality monitoring is essential due to its significant impact on human health and the environment. A hybrid AQI classification model is presented in this study, combining the use of deep learning-based features (VGG16), classical image processing techniques (including color and texture analysis) and advanced outlier detection methods (DBSCAN, LOF, and Isolation Forest). The proposed approach not only improves the accuracy of the classification but it also improves the detection of anomalies, especially in the case of extreme pollution, addressing the key limitations of existing image-based AQI models. In the experimental evaluation, 96% accuracy was achieved on the original dataset and 77% on the unseen dataset, and the drop in accuracy is attributed to the limited size of the unseen dataset, affecting generalization performance. In-spite of this, it shows significant robust-ness under varied environmental setups. Our findings show that this scalable and reliable framework for real-time AQI classification can bring a more accurate air quality assessment that supports initiatives in public health and environmental protection.