Development of Artificial Intelligent-Based Methodology to Prepare Input for Estimating Vehicle Emissions DOI Creative Commons
Elif Yavuz, Asım Öztürk, Nedime Gaye Nur Balkanlı

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(23), P. 11175 - 11175

Published: Nov. 29, 2024

Machine learning has significantly advanced traffic surveillance and management, with YOLO (You Only Look Once) being a prominent Convolutional Neural Network (CNN) algorithm for vehicle detection. This study utilizes version 7 (YOLOv7) combined the Kalman-based SORT (Simple Online Real-time Tracking) as one of models used in our experiments real-time identification. We developed “ISTraffic” dataset. have also included an overview existing datasets domain detection, highlighting their shortcomings: detection often incomplete annotations limited diversity, but dataset addresses these issues detailed extensive higher accuracy robustness. The ISTraffic is meticulously annotated, ensuring high-quality labels every visible object, including those that are truncated, obscured, or extremely small. With 36,841 annotated examples average 32.7 per image, it offers coverage dense annotations, making highly valuable various object tracking applications. enhance capabilities, enabling development more accurate reliable complex environments. comprehensive versatile, suitable applications ranging from autonomous driving to surveillance, improved performance, resulting robustness challenging scenarios. Using this dataset, achieved significant results YOLOv7 model. model demonstrated high detecting types, even under conditions. highlight effectiveness training robust underscore its potential future research field. Our comparative analysis evaluated against variants, YOLOv7x YOLOv7-tiny, using both COCO (Common Objects Context) benchmark. outperformed others [email protected] 0.87, precision 0.89, recall 0.84, showing 35% performance improvement over COCO. Performance varied different conditions, daytime yielding compared night-time rainy weather, where headlights affected contours. Despite effective counting, high-speed vehicles remains challenge. Additionally, algorithm’s deep estimates emissions (CO, NO, NO2, NOx, PM2.5, PM10) were 7.7% 10.1% lower than ground-truth.

Language: Английский

One-Point Calibration of Low-Cost Sensors for Particulate Air Matter (PM) Concentration Measurement DOI Creative Commons
Luigi Russi, Roberto Guidorzi, Giovanni Semprini

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(3), P. 692 - 692

Published: Jan. 24, 2025

The use of low-cost sensors has dramatically increased in recent years all engineering sectors. In the buildings and automotive field, open very interesting perspectives, because they allow one to monitor temperature humidity distributions together with air quality a widespread punctual way for control energy parameters. main issue remains validation measurements. this work, we propose an innovative approach verify measurements given by some systems built ad hoc applications. Two independent measurement were set measure Particulate Air Matter (PM) concentration, TVOC CO2 formaldehyde temperature, relative humidity, pressure, flow velocity, GPS position. These calibrated PM concentration comparison standard certified used regional authority Emilia-Romagna region (ARPAE, Italy) characterizing quality. duration analysis, three days, is not representative diverse environmental conditions that occur across different seasons. However, innovation lies both in-field high-quality proper conversion approaches mass A quantitative analysis sensors’ performance given, focus on effects time granularity, from particle counts, size detection response. results show number are good agreement measurements, strong impact indicators. Overall, consistency data among achieved.

Language: Английский

Citations

0
Low-Cost Particulate Matter Mass Sensors: Review of the Status, Challenges, and Opportunities for Single-Instrument and Network Calibration DOI
Jingzhuo Zhang, Li Bai, Na Li

et al.

ACS Sensors, Journal Year: 2025, Volume and Issue: unknown

Published: May 7, 2025

As an emerging atmospheric monitoring technology, low-cost sensors for particulate matter of diameters below 2.5 μm (PM2.5LCSs) supplement traditional air quality instruments. Because their stability and accuracy are typically low, they require adequate calibration to meet operational requirements. Numerous studies have now been published on single-sensor PM2.5LCS models, research networks, designed measure pollutant concentration with high spatiotemporal resolution, is gradually starting. However, there no established standard procedure sensor calibration. Here we comprehensively reviewed evaluate the current status, identify major challenges, provide support applications networks. Regression machine learning were most common methods single PM2.5LCSs. Environmental factors duration period influenced model accuracy, especially (data-driven) algorithms. For included early evaluation homogeneous or colocated Method selection depended regional environmental conditions, concentration, presence absence reference Quality control crucial operation network, online drift detection management measures routine assurance control. In conclusion, use, intensive machine-learning-based must be conducted practical application large-scale

Language: Английский

Citations

0
Machine Learning–Based Calibration and Performance Evaluation of Low-Cost Internet of Things Air Quality Sensors DOI Creative Commons
Mehmet Taştan

Sensors, Journal Year: 2025, Volume and Issue: 25(10), P. 3183 - 3183

Published: May 19, 2025

Low-cost air quality sensors (LCSs) are increasingly being used in environmental monitoring due to their affordability and portability. However, sensitivity factors can lead measurement inaccuracies, necessitating effective calibration methods enhance reliability. In this study, an Internet of Things (IoT)-based system was developed tested using the most commonly preferred sensor types for measurement: fine particulate matter (PM2.5), carbon dioxide (CO2), temperature, humidity sensors. To improve accuracy, eight different machine learning (ML) algorithms were applied: Decision Tree (DT), Linear Regression (LR), Random Forest (RF), k-Nearest Neighbors (kNN), AdaBoost (AB), Gradient Boosting (GB), Support Vector Machines (SVM), Stochastic Descent (SGD). Sensor performance evaluated by comparing measurements with a reference device, best-performing ML model determined each sensor. The results indicate that GB kNN achieved highest accuracy. For CO2 calibration, R2 = 0.970, RMSE 0.442, MAE 0.282, providing lowest error rates. PM2.5 sensor, delivered successful results, 2.123, 0.842. Additionally, temperature sensors, demonstrated accuracy values (R2 0.976, 2.284). These findings demonstrate that, identifying suitable methods, ML-based techniques significantly LCSs. Consequently, they offer viable cost-effective alternative traditional high-cost systems. Future studies should focus on long-term data collection, testing under diverse conditions, integrating additional further advance field.

Language: Английский

Citations

0
Assessment of vertical transport of PM in a surface iron ore mine due to in-pit mining operations DOI

Abhishek Penchala,

Aditya Kumar Patra, Samrat Santra

et al.

Measurement, Journal Year: 2024, Volume and Issue: 240, P. 115580 - 115580

Published: Aug. 24, 2024

Language: Английский

Citations

3
Design and Enhancement of a Fog-Enabled Air Quality Monitoring and Prediction System: An Optimized Lightweight Deep Learning Model for a Smart Fog Environmental Gateway DOI Creative Commons

P. Divya Bharathi,

V. Anantha Narayanan,

P. Bagavathi Sivakumar

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(15), P. 5069 - 5069

Published: Aug. 5, 2024

Effective air quality monitoring and forecasting are essential for safeguarding public health, protecting the environment, promoting sustainable development in smart cities. Conventional systems cloud-based, incur high costs, lack accurate Deep Learning (DL)models multi-step forecasting, fail to optimize DL models fog nodes. To address these challenges, this paper proposes a Fog-enabled Air Quality Monitoring Prediction (FAQMP) system by integrating Internet of Things (IoT), Fog Computing (FC), Low-Power Wide-Area Networks (LPWANs), (DL) improved accuracy efficiency levels. The three-layered FAQMP includes low-cost (AQM) node transmitting data via LoRa layer then cloud complex processing. Smart Environmental Gateway (SFEG) FC introduces efficient Intelligence employing an optimized lightweight DL-based Sequence-to-Sequence (Seq2Seq) Gated Recurrent Unit (GRU) attention model, enabling real-time processing, timely warnings dangerous AQI levels while optimizing resource usage. Initially, Seq2Seq GRU Attention validated outperformed state-of-the-art methods with average RMSE 5.5576, MAE 3.4975, MAPE 19.1991%, R

Language: Английский

Citations

2
Optimisation of the adaptive neuro-fuzzy inference system for adjusting low-cost sensors PM concentrations DOI Creative Commons
Martina Casari, Piotr A. Kowalski, Laura Po

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 83, P. 102781 - 102781

Published: Aug. 23, 2024

Language: Английский

Citations

0
Development of Artificial Intelligent-Based Methodology to Prepare Input for Estimating Vehicle Emissions DOI Creative Commons
Elif Yavuz, Asım Öztürk, Nedime Gaye Nur Balkanlı

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(23), P. 11175 - 11175

Published: Nov. 29, 2024

Machine learning has significantly advanced traffic surveillance and management, with YOLO (You Only Look Once) being a prominent Convolutional Neural Network (CNN) algorithm for vehicle detection. This study utilizes version 7 (YOLOv7) combined the Kalman-based SORT (Simple Online Real-time Tracking) as one of models used in our experiments real-time identification. We developed “ISTraffic” dataset. have also included an overview existing datasets domain detection, highlighting their shortcomings: detection often incomplete annotations limited diversity, but dataset addresses these issues detailed extensive higher accuracy robustness. The ISTraffic is meticulously annotated, ensuring high-quality labels every visible object, including those that are truncated, obscured, or extremely small. With 36,841 annotated examples average 32.7 per image, it offers coverage dense annotations, making highly valuable various object tracking applications. enhance capabilities, enabling development more accurate reliable complex environments. comprehensive versatile, suitable applications ranging from autonomous driving to surveillance, improved performance, resulting robustness challenging scenarios. Using this dataset, achieved significant results YOLOv7 model. model demonstrated high detecting types, even under conditions. highlight effectiveness training robust underscore its potential future research field. Our comparative analysis evaluated against variants, YOLOv7x YOLOv7-tiny, using both COCO (Common Objects Context) benchmark. outperformed others [email protected] 0.87, precision 0.89, recall 0.84, showing 35% performance improvement over COCO. Performance varied different conditions, daytime yielding compared night-time rainy weather, where headlights affected contours. Despite effective counting, high-speed vehicles remains challenge. Additionally, algorithm’s deep estimates emissions (CO, NO, NO2, NOx, PM2.5, PM10) were 7.7% 10.1% lower than ground-truth.

Language: Английский

Citations

0