A Survey of Methods and Architectures for Enhancing Air Pollution Detection Accuracy and Quality Monitoring Using Pyramidal Convolution Split-Attention Networks and IoT

Abstract

Air pollution monitoring and prediction are critical for mitigating environmental and public health risks. Traditional air quality monitoring systems rely on fixed stations, which are costly and provide limited spatial coverage. The integration of Artificial Intelligence (AI) and Internet of Things (IoT) technologies has significantly improved the efficiency and accuracy of air pollution detection systems. IoT-based sensor networks enable real-time collection of environmental data, while AI models process this data to identify complex spatiotemporal patterns. Recent advances in deep learning, particularly hybrid architectures such as CNN-LSTM and attention-based models, have demonstrated superior performance in air quality prediction tasks. Studies show that deep learning models outperform traditional machine learning approaches by effectively handling nonlinear and time-series data. Furthermore, advanced architectures such as pyramidal convolution and split-attention networks enhance multi-scale feature extraction and adaptive learning, improving prediction accuracy. This paper presents a comprehensive survey of AI-based IoT air pollution monitoring systems from 2020 to 2023. It analyses methodologies, comparative performance, and challenges, including computational complexity, data quality, and scalability. The study also identifies future research directions toward efficient and interpretable air quality monitoring systems.