A Comprehensive Review of Enhancing Air Pollution Detection Accuracy and Quality Monitoring Using Pyramidal Convolution Split-Attention Networks and IoT

Abstract

Air pollution has become a major environmental challenge affecting human health, climate stability, and ecological sustainability. Accurate monitoring of pollutants such as PM2.5, CO₂, NO₂, and SO₂ is essential for environmental protection and effective decision-making. Traditional air quality monitoring systems are often expensive, limited in coverage, and unable to provide real-time analysis. The integration of Internet of Things (IoT) technologies with deep learning models has significantly improved air pollution detection by enabling continuous, distributed, and cost-effective environmental monitoring. IoT sensor networks collect real-time atmospheric data, while deep learning architectures such as CNNs, LSTMs, and attention-based models analyze complex pollutant patterns and improve prediction accuracy. Advanced architectures including pyramidal convolution networks, split-attention mechanisms, and hybrid CNN-BiLSTM frameworks have demonstrated strong performance in Air Quality Index prediction and pollution classification tasks. In addition, optimization techniques such as feature engineering, dimensionality reduction, and hyperparameter tuning enhance computational efficiency and model reliability. Attention-based and multi-scale deep learning models further improve feature extraction from heterogeneous IoT data streams. Despite these advancements, challenges including sensor calibration, scalability, data heterogeneity, and energy efficiency remain significant concerns. Emerging approaches such as federated learning, edge computing, and graph-based deep learning offer promising directions for future intelligent air quality monitoring systems.