Deep Learning and Optimization Approaches in Enhancing Air Pollution Detection Accuracy and Quality Monitoring Using Pyramidal Convolution Split-Attention Networks and IoT: A Review

Abstract

Air pollution has emerged as one of the most critical environmental and public health challenges worldwide. The increasing concentration of harmful pollutants such as PM2.5, CO₂, NO₂, and SO₂ has necessitated the development of accurate and real-time monitoring systems. Traditional air quality monitoring methods suffer from limited spatial coverage, delayed data processing, and low predictive accuracy. Recent advancements in Internet of Things (IoT), machine learning (ML), and deep learning (DL) have enabled the development of intelligent air quality monitoring systems capable of real-time data acquisition and prediction. This review focuses on deep learning and optimization approaches for enhancing air pollution detection accuracy using IoT-enabled systems and advanced architectures such as pyramidal convolution and split-attention networks. Recent studies indicate that deep learning models, especially CNN, LSTM, and hybrid architectures, significantly improve prediction accuracy by capturing complex spatio-temporal relationships in air quality data. Furthermore, IoT-based sensor networks enable continuous monitoring and data collection, which enhances model performance and real-time decision-making. The paper provides a systematic review of studies from 2020 to 2023, highlighting key architectures, methodologies, and optimization techniques. A comparative analysis is conducted based on accuracy, computational efficiency, and scalability. Finally, the paper discusses challenges such as data heterogeneity, sensor reliability, and energy consumption, and suggests future research directions including edge intelligence, attention-based architectures, and hybrid deep learning models.