Environmental Image-Based PM2.5 Prediction Using Vision Transformers

Abdullah Burhan; Yasir Saleem Afridi; Bilal Ur Rehman; Kifayat Ullah; Wasim Habib; Muhammad Iftikhar Khan; Muhammad Amir

Authors

Abdullah Burhan University of Engineering and Technology, Peshawar, Pakistan
Yasir Saleem Afridi University of Engineering and Technology, Peshawar, Pakistan
Bilal Ur Rehman University of Engineering and Technology, Peshawar, KPK, Pakistan
Kifayat Ullah University of Engineering and Technology, Peshawar, Pakistan
Wasim Habib University of Engineering and Technology, Peshawar, Pakistan
Muhammad Iftikhar Khan University of Engineering and Technology, Peshawar, Pakistan
Muhammad Amir University of Engineering and Technology, Peshawar, Pakistan

Keywords:

PM2.5 Prediction, Vision Transformer, Air Quality Monitoring, Environmental Image Analysis, Self-Attention, Deep Learning Regression

Abstract

Fine particulate matter (PM2.5) is one of the most harmful atmospheric pollutants, posing significant risks to human health and environmental sustainability. Conventional PM2.5 monitoring systems rely on dense sensor networks, which are often expensive, require regular maintenance, and provide limited spatial coverage, particularly in resource-constrained regions. This study proposes an image-based PM2.5 prediction framework the proposed Vision Transformer framework achieves to estimate PM2.5 concentrations directly from environmental RGB images without depending solely on physical sensors. A dataset comprising 6,587 environmental images from eight geographic classes was synchronized with ground-truth PM2.5 measurements and processed through image resizing, normalization, illumination correction, and data augmentation. To improve model interpretability and robustness, physics-informed visual descriptors, including atmospheric transmission, sky smoothness, color ratio, contrast, entropy, and solar zenith angle, were integrated with transformer-based patch embeddings. The proposed ViT model utilizes multi-head self-attention to capture long-range spatial dependencies associated with atmospheric haze and scattering effects, which are difficult to model using conventional convolutional architectures. The performance of the proposed framework was evaluated against Support Vector Regression (SVR), Random Forest (RF), and CNN-LSTM models using regression-based metrics, including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). The results demonstrate that the proposed Vision Transformer framework achieves superior PM2.5 prediction performance, showing improved accuracy and robustness across urban and semi-urban environments and varying atmospheric conditions. The findings indicate that environmental images contain valuable visual information for reliable PM2.5 estimation, providing a low-cost, scalable, and sensor-independent solution for future air quality monitoring applications, particularly in areas with limited monitoring infrastructure.

References

G. Wang et al., “A Prototype Feature Representation-Guided Multi-Task Learning Framework for Image-based PM2:5Concentration Monitoring,” IEEE Trans. Geosci. Remote Sens., 2026, doi: 10.1109/TGRS.2026.3704231.

Hirokazu Madokoro, Stephanie Nix, “Multimodal Particulate Matter Prediction: Enabling Scalable and High-Precision Air Quality Monitoring Using Mobile Devices and Deep Learning Models,” Sensors, vol. 25, no. 13, 2025, doi: https://doi.org/10.3390/s25134053.

“PM25Vision: A Large-Scale Benchmark Dataset for Visual Estimation of Air Quality Yang Han,” arXiv:2509.16519, 2025, [Online]. Available: https://arxiv.org/abs/2509.16519

M. P. Vafa, “Image-Based Air Quality Estimation with Deep Learning: A Systematic Review,” May 2026, doi: 10.20944/PREPRINTS202604.2212.V1.

Joyanta Jyoti Mondal, Md. Farhadul Islam, Raima Islam, “Uncovering local aggregated air quality index with smartphone captured images leveraging efficient deep convolutional neural network,” Sci. Rep., 2024, [Online]. Available: https://www.nature.com/articles/s41598-023-51015-1

K. Gu et al., “Air Pollution Monitoring by Integrating Local and Global Information in Self-Adaptive Multiscale Transform Domain,” IEEE Trans. Multimed., vol. 27, pp. 3716–3728, 2025, doi: 10.1109/TMM.2025.3535351.

Om Kathalkar, Nitin Nilesh, Sachin Chaudhari, Anoop Namboodiri, “AQIFormer: A Transformer-Based Multi-View Architecture for Cross-City Air Quality Classification,” arXiv:2606.07648, 2026, [Online]. Available: https://arxiv.org/abs/2606.07648

Mohammad Saleh Vahdatpour, Maryam Eyvazi, Yanqing Zhang, “Forecasting and Visualizing Air Quality from Sky Images with Vision-Language Models,” arXiv:2509.15076, 2025, [Online]. Available: https://arxiv.org/abs/2509.15076

“(PDF) Vision-Based Estimation of PM2.5 from Surveillance Images.” Accessed: Jul. 05, 2026. [Online]. Available: https://www.researchgate.net/publication/405480687_Vision-Based_Estimation_of_PM25_from_Surveillance_Images

C. Y. Lee and L. Moses Manova, “Enhancing PM2.5 Detection via Multispectral Image-Based Double-Channel Ensemble Learning for Drones,” IEEE Sens. J., vol. 25, no. 17, pp. 33528–33539, 2025, doi: 10.1109/JSEN.2025.3585177.

M. A. Xiang Zhang, “RTEMSF-Mamba: Real-time estimation and short-term multi-step forecasting of PM2.5 concentrations using spatiotemporal deep learning,” Build. Environ., vol. 295, p. 114449, 2026, doi: https://doi.org/10.1016/j.buildenv.2026.114449.

Mohamed Mokhtar Ouardi, Dayang N. A. Jawawi, “Enhancing Vision Transformer with Vision language Model Text Embeddings for Robust Air Pollution Classification,” Int. J. Innov. Comput., vol. 16, no. 1, pp. 97–108, 2026, doi: 10.11113/ijic.v16n1.678.

Yash Mishra, Kedarnath senapati, “DeepAQI: A Vision-Based EfficientNet Framework for Air Quality Index Prediction from Environmental Metadata,” Researchgate, 2026, doi: 10.21203/rs.3.rs-9062054/v1.

B. U. Patil, R. Chetan, and V. Harshitha, “Explainable multimodal deep learning system for sustainable urban air quality forecasting and risk-aware recommendations,” Model. Earth Syst. Environ. 2026 122, vol. 12, no. 2, pp. 123-, Mar. 2026, doi: 10.1007/S40808-026-02771-2.

Y. Mehta, V. Kamath, N. Kini, A. Haraniya, and S. Agrawal, “Estimating Ground-Level Air Quality Index from Satellite Imagery: A Data-Driven Approach,” 2025 2nd IEEE Int. Conf. Women Comput. InCoWoCo 2025, 2025, doi: 10.1109/INCOWOCO68239.2025.11407074.

T. Srimanee, B. Kiatphibun, C. Thanomngern, N. Mhuaksa, C. Banju-ngam, and T. Siriborvornratanakul, “Multimodal PM2.5 Forecasting Using Satellite Imagery and Sensor Data with Semi-supervised Deep Learning,” Earth Syst. Environ. 2026, pp. 1–18, Apr. 2026, doi: 10.1007/S41748-026-01109-3.

Z. Zhang & S. Zhang, “Modeling air quality PM2.5 forecasting using deep sparse attention-based transformer networks,” Int. J. Environ. Sci. Technol., vol. 20, pp. 13535–13550, 2023, [Online]. Available: https://link.springer.com/article/10.1007/s13762-023-04900-1

S. Lee and S. Sim, “Spatio-Temporal Multimodal Large Language Model for Air Quality Index Prediction,” Int. Conf. Electr. Comput. Energy Technol. ICECET 2025, 2025, doi: 10.1109/ICECET63943.2025.11472437.

Israt Tabassum, G. M. Ansarul Kabir, “A deep learning approach for multi-class classification of air quality from image data,” Comput. Sci. Eng. Res., 2025, doi: 10.69517/cser.2025.02.01.0003.

M. Wang, C. Bi, L. Yang, X. Qiu, Y. Li, and C. Yu, “PMIM: generating high-resolution air pollution data via masked image modeling,” J. Vis. 2024 273, vol. 27, no. 3, pp. 383–399, Mar. 2024, doi: 10.1007/S12650-024-00965-3.

Z. Wang, Y. Yang, and S. Yue, “Air Quality Classification and Measurement Based on Double Output Vision Transformer,” IEEE Internet Things J., vol. 9, no. 21, pp. 20975–20984, Nov. 2022, doi: 10.1109/JIOT.2022.3176126.

H. Yagnik, R. K. Gupta, and A. Pandya, “Deep Learning for Air Pollution Prediction: A Systematic Review of CNN, LSTM, GNN, and Transformer Architectures,” Arch. Comput. Methods Eng. 2026, pp. 1–26, Jun. 2026, doi: 10.1007/S11831-026-10639-Y.

P. J. Ramal and E. Anbalagan, “Hybrid Swin Transformer-BiGRU Framework for Spatio-Temporal Prediction of Urban Air, Water, and Waste Dynamics in Smart Cities,” Proc. 3rd Int. Conf. Sustain. Comput. Data Commun. Syst. ICSCDS 2025, pp. 1525–1535, 2025, doi: 10.1109/ICSCDS65426.2025.11167444.