PPDRTL:A Novel framework for Predicting Pollutant Deposition on Roadside Tree Leaves Using Linear Regression

PPDRTL DATASET MATLAB README.md This file was generated on 2026-04-07 for the PPDRTL framework. GENERAL INFORMATION Title of Dataset: PPDRTL:A Novel framework for Predicting Pollutant Deposition on Roadside Tree Leaves Using Linear Regression Geographic Location of Data Collection: Gurugram, Haryana, India (Rajeev Chowk, Sheetla Mata Road, and Sector 12). DATA & FILE OVERVIEW File List: A) pso.m: The primary execution script for the PPDRTL framework. Implements multi-resolution denoising (DWT/SWT), metaheuristic segmentation (PSO/DPSO/FODPSO), and linear regression. B) AQI data.xlsx: Ground-truth air quality data containing PM2.5, SO2, NO2, CO, and O3 measurements from the Vikas Sadan station. C) whorled leaf.xlsx / Leafimage.xlsx: Datasets mapping leaf characteristics and species morphology to their respective pollution capture metrics. Relationship Between Files: AQI data.xlsx serves as the reference ground truth for the predictive model (pso.m). The image datasets provide the input features extracted during the digital number (DN) extraction phase. LEAF MORPHOLOGY & ARRANGEMENT The framework analyzes specific species known for their particulate capture efficiency. Whorled leaf.zip: Whorled Leaf Arrangement (High/Low Traffic) Species: Alstonia scholaris and Alstonia macrophylla. Opposite leaf.zip: Opposite Leaf Arrangement (Medium Traffic) Species: Millettia pinnata. Alternate leaf.zip: Alternate Leaf Arrangement (Bio-indicators) Species: Polyalthia longifolia, Azadirachta indica. METHODOLOGICAL INFORMATION Description of Methods: The PPDRTL framework follows a four-stage signal processing and machine learning pipeline: Wavelet Preprocessing: Applies DWT and SWT for denoising while preserving edge texture. Threshold Segmentation: Optimizes Otsu’s variance criterion using metaheuristic algorithms (PSO, DPSO, FODPSO). DN Feature Extraction: Quantifies Contrast, Entropy, and Standard Deviation from segmented regions. Predictive Modeling: Maps extracted features to PM2.5 and AQI values using Linear Regression or Neural Networks. Software Requirements: MATLAB (R2021b or later). Required Toolboxes: Image Processing, Wavelet, Deep Learning, Statistics & Machine Learning. EXPERIMENTAL RESULTS The following performance was validated during the 10-day Gurugram pilot study: Traffic Density Segmentation Method Avg. PM2.5 Accuracy (%) R² Score High FODPSO 88.91% 0.856 Medium FODPSO 90.70% 0.894 Low FODPSO 91.42% 0.912 USER MANUAL (HOW TO RUN) This section provides a detailed guide on setting up the environment, organizing datasets, and executing the PPDRTL framework. 1. System & Hardware Requirements Operating System: Windows / Linux / macOS (tested on R2021b). Memory: 16GB RAM minimum. GPU Acceleration: Supported via CUDA-enabled NVIDIA GPUs (e.g., RTX 3060+ with 8GB VRAM) for accelerated Deep Learning training. MATLAB Toolboxes: Image Processing, Wavelet, Deep Learning, Statistics and Machine Learning Toolbox. 2. Project Directory Structure Maintain the following structure for automated pathing to function correctly: /Project Root/ ├── pso.m # Swarm-optimized pipeline (Primary) ├── comparemodels.m # Visualization utilities ├── AQI data.xlsx # High Traffic (HT) reference ├── AQI data LT.xlsx # Low Traffic (LT) reference ├── data AQI.xlsx # AQI data of whorled leaf reference └── /Leafimage/ # Raw RGB sample repository 3. Data Integration & Preparation Standard Metadata: Ensure AQI data.xlsx has columns for Date, PM2.5, and matching ImageName. 4. Step-by-Step Execution Workflow Phase A: Core Pipeline (Baseline) Open MATLAB and navigate to the project root. Open pso.m. Set the useSwarm flag (1: PSO, 2: DPSO, 3: FODPSO). Run the script. Results will output the predicted PM2.5 and calculated DN features in the command window. 5. Advanced Configuration Swarm Density: Modify Np (default 30) in Algorithm 2 section of pso.m to adjust search space coverage. 6. Troubleshooting Error: "Out of Memory": Close other MATLAB instances. Path Error: Verify that datasetFolder in the script headers matches your actual absolute path (e.g., h:/MATLAB DRIVE/Leafimage/). ADVANCED RESEARCH PIPELINE (CORE COMPONENTS) This section details the finalized algorithmic frameworks and datasets used for the PPDRTL study, focusing on swarm-optimized segmentation and multimodal fusion. 1. Component Interaction Flowchart ┌─────────────────────┐ ┌─────────────────────────────────┐ │ Raw Leaf Images │ │ AQI data.xlsx (High Traffic) │ └────────┬────────────┘ │ AQI data LT.xlsx (Low Traffic) │ │ └───────────────┬─────────────────┘ │ pso.m: DWT/SWT │ │ Wavelet Denoising │ Ground Truth Labels ▼ │ ┌─────────────────────┐ │ │ Denoised Image │ │ └────────┬────────────┘ │ │ │ │ Swarm Optimization │ │ (PSO / DPSO / FODPSO) │ ▼ │ ┌─────────────────────┐ │ │ Segmented Mask │ │ └────────┬────────────┘ │ │ │ │ Feature Extraction │ ▼ │ ┌──────────────────────────────┐ │ │ DN Features: │ │ │ Contrast, Entropy, StdDev │ │ └───────┬──────────┬───────────┘ │ │ │ │ Linear│ MLP │ │ Regression Regressor │ │ │ │ ▼ ▼ │ ┌──────────┐ ┌──────────────┐ │ │ pso.m │ │predict pollut│ │ │Prediction│ │ant final.m │ │ └────┬─────┘ └──────┬───────┘ │ │ │ │ └──────┬───────┘ │ │ │ │ Performance Metrics │ ▼ │ ┌───────────────────────────────────────────┐ │ comparemodels.m │◄─── Ground Truth │ compare_pollutant_models_5fold.m │ (Validation) │ (R², RMSE, MAE — 5-Fold Benchmarking) │ └───────────────────────────────────────────┘ 2. Detailed File Descriptions A) Excel Datasets (Ground Truth) AQI data.xlsx: Primary dataset containing 10-day monitoring records for PM2.5, NO2, SO2, CO, and O3. Represents baseline high-traffic urban conditions. AQI data LT.xlsx: Specialized dataset for Low Traffic (LT) scenarios. Crucial for understanding how leaf capture efficiency differs in less-polluted environments. NISE Gwal Pahari Gurugram-IMD.xlsx: Validation data from the National Institute of Solar Energy (NISE) station, providing a cross-reference for predicted atmospheric concentrations. B) MATLAB Algorithms (.m files) pso.m: The primary research pipeline. It combines DWT/SWT wavelet denoising with metaheuristic segmentation. The script allows switching between standard PSO, Diversity-enhanced DPSO, and Fractional-order FODPSO for optimal leaf thresholding. compare pollutant models 5fold.m: A rigorous benchmarking script that uses 5-fold cross-validation to compare Decision Trees, Random Forests, SVMs, and customized Neural Networks. predict pollutant final.m: A finalized regression script that utilizes a deep-learning architecture (without transformer layers) to map leaf textures directly to O3 or PM2.5 concentrations. comparemodels.m: A visualization utility that generates the comparative scatter plots and computes R², RMSE, and MAE metrics for model performance analysis. DATA-SPECIFIC INFORMATION AQI data.xlsx Number of Variables: 7 Variable List: Date, Station (Vikas Sadan), PM2.5 (µg/m³), NO2 (ppb), SO2 (ppb), CO (mg/m³), O3 (ppb). pso.m Results Number of Variables: 5 Variable List: Contrast, Entropy, Standard Deviation, Prediction (forecasted PM2.5), Error. AUTHOR INFORMATION Authors: Name Affiliation Email Neelam Yadav‡ Dept. of CSE, Chandigarh College of Engineering and Technology, India [e-mail] Sunil K. Singh‡ Dept. of CSE, Chandigarh College of Engineering and Technology, India [e-mail] Dinesh Sharma‡ Dept. of ECE, Chandigarh College of Engineering and Technology, India [e-mail] Sudhakar Kumar‡ Dept. of CSE, Chandigarh College of Engineering and Technology, India [e-mail] Varsha Arya† Hong Kong Metropolitan University (HKMU), Hong Kong, China [e-mail] Wadee Alhalabi Dept. of CS, King Abdulaziz University, Jeddah, Saudi Arabia [e-mail] Hind Alsharif Dept. of CS, College of Computing, Umm Al-Qura University, Saudi Arabia [e-mail] Brij B. Gupta* Dept. of CS & Info. Engg., Asia University, Taiwan [e-mail] Contact: For questions or collaboration, please contact the corresponding author: Dr. Sudhakar Kumar – [e-mail]