Przegląd Geograficzny (2025) vol. 97, iss. 1
Intelligent air pollution identification system
Przegląd Geograficzny (2025) vol. 97, iss. 1, pp. 49-68 | 
Full text
doi: https://doi.org/10.7163/PrzG.2025.1.3
Abstract
The primary objective of this study is to develop an intelligent system for air pollution identification using neural networks. The implementation of artificial intelligence (AI), leveraging the analysis of surface images of selected air pollution indicators, has enabled the creation of a low‑cost‑effective and efficient method for detecting hazardous substances.
The study employs scanning electron microscopy (SEM) images of two‑year‑old Scots pine needles (Pinus sylvestris L.) collected from representative research catchments of the national Integrated Environmental Monitoring network. These images serve as input data for a machine learning algorithm designed to classify and segment air pollution particles based on predefined attributes such as size, shape, and chemical composition. To prepare the data for machine learning, SEM images were processed using graphic software, where classified particles were assigned distinct colours, corresponding to their characteristics. The processed layers, referred to as masks, became essential components in training a deep learning model to automatically recognize and categorize pollutants.
The key innovation of this study lies in the use of a self‑learning algorithm, which optimizes the analysis of contaminants deposited on the surface of pine needles, offering a reliable approach to air pollution assessment without the need for costly traditional measurement devices. The neural network employed in this study was structured with multiple convolutional layers, allowing it to capture intricate details within the images during training. These layers extract essential features from the images and apply them to generate segmentation masks that highlight the presence of pollution particles. By associating input pixels with their corresponding features and comparing them to pre‑labeled masks, the model continuously adjusts its parameters, thereby improving its predictive accuracy over time. The system achieved an 80% prediction accuracy, demonstrating its potential as a reliable tool for identifying air pollutants.
The analysis of pine needles serves as a bioindicator‑based approach to monitoring air pollution, complementing existing measurement techniques. The pine needles, sampled from various locations experiencing different levels of anthropogenic pressure, provide insight into the regional variations in pollution levels. The chemical composition of the particles adhered to the needles was analysed using Energy Dispersive Spectroscopy (EDS), which allowed for a more detailed understanding of the pollutants’ origins.
The study identified key contaminants such as mineral dust, industrial particulate matter, metallic spherules, and fine atmospheric aerosols. One of the most significant findings of this research is the ability of the AI‑powered system to differentiate between natural and anthropogenic pollution sources. The model was trained to recognize four distinct categories of pollution, each represented by a specific colour in the processed masks: green for stomatal surfaces, yellow for sharp‑edged industrial particles, red for magnetic spherules, and blue for fine mineral dust. This classification system ensures precise pollutant identification, which is critical for environmental monitoring and policy‑making. The integration of deep learning techniques into environmental studies marks a shift toward more advanced and automated pollution assessment methods. Traditional air quality monitoring relies on complex and expensive measurement infrastructure, whereas the proposed AI‑based approach provides a low‑cost and scalable alternative. The ability of the system to analyse new datasets without requiring manual preprocessing highlights its practical applicability for real‑world environmental monitoring. In future research, additional efforts will focus on refining the model to further improve its accuracy and expand its application to other vegetation types and environmental indicators. The next phase of development will also explore the quantification of specific pollutants within each identified category, enabling a more precise determination of pollution sources. By incorporating additional datasets and refining neural network architectures, this AI-driven approach could play a vital role in global air quality monitoring initiatives.
Keywords: air pollution identification, artificial intelligence, machine learning, electron microscopy, neutral ne- tworks, image analysis
[miroslaw.szwed@ujk.edu.pl], Uniwersytet Jana Kochanowskiego w Kielcach, Instytut Geografii i Nauk o Środowisku
Citation
APA: Szwed, M. (2025). Inteligentny system identyfikacji zanieczyszczenia powietrza. Przegląd Geograficzny, 97(1), 49-68. https://doi.org/10.7163/PrzG.2025.1.3
MLA: Szwed, Mirosław. "Inteligentny system identyfikacji zanieczyszczenia powietrza". Przegląd Geograficzny, vol. 97, no. 1, 2025, pp. 49-68. https://doi.org/10.7163/PrzG.2025.1.3
Chicago: Szwed, Mirosław. "Inteligentny system identyfikacji zanieczyszczenia powietrza". Przegląd Geograficzny 97, no. 1 (2025): 49-68. https://doi.org/10.7163/PrzG.2025.1.3
Harvard: Szwed, M. 2025. "Inteligentny system identyfikacji zanieczyszczenia powietrza". Przegląd Geograficzny, vol. 97, no. 1, pp. 49-68. https://doi.org/10.7163/PrzG.2025.1.3