🤖 machine learning

A Scoping Review of Deep Learning for Urban Visual Pollution and Proposal of a Real-Time Monitoring Framework with a Visual Pollution Index

This scoping review analyzes existing deep learning methods for detecting urban visual pollution and proposes a real-time monitoring framework featuring a visual pollution index to support standardized and sustainable urban management.

Original authors: Mohammad Masudur Rahman, Md. Rashedur Rahman, Ashraful Islam, Saadia B Alam, M Ashraful Amin

Published 2026-02-11

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Mohammad Masudur Rahman, Md. Rashedur Rahman, Ashraful Islam, Saadia B Alam, M Ashraful Amin

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are walking through a beautiful, historic city. Suddenly, your eyes are hit by a massive, neon-colored billboard blocking a view of a cathedral, a tangle of messy black wires crisscrossing the sky like a spiderweb, and piles of trash overflowing on a sidewalk.

Even though this doesn't make you physically sick like smog or dirty water, it makes you feel stressed, tired, and frustrated. This is Urban Visual Pollution (UVP)—the "eye sore" of the modern world.

This research paper is essentially a "Blueprint for a Digital City Janitor." Here is the breakdown of what the researchers did and what they are proposing.

1. The "Library Search" (The Scoping Review)

First, the researchers acted like detectives. They searched through massive digital libraries to see what other scientists have already done to fight visual pollution.

They found that while people are using "smart" technology (Deep Learning/AI) to spot things like trash or messy signs, the work is currently fragmented. It’s like having a dozen different people trying to clean a house, but one person only cleans the windows, another only sweeps the floor, and nobody is talking to each other. There is no master plan.

2. The "Brain" of the System (Deep Learning)

The paper discusses using Deep Learning, which you can think of as teaching a computer to have "eyes" and a "brain."

The Eyes (Object Detection): Using models like "YOLO" (which stands for You Only Look Once), the computer can scan a video feed and instantly point a finger at a piece of litter or a broken sign.
The Brain (Classification): The computer doesn't just see "a thing"; it understands what the thing is. It can tell the difference between a permitted shop sign and an illegal, messy poster.

3. The "Visual Pollution Index" (The Fever Thermometer)

This is one of the most important parts of their proposal. Currently, cities don't have a way to "measure" how ugly or chaotic a street is.

The researchers propose a Visual Pollution Index (VPI). Think of this like a fever thermometer for a city street.

If a street has a little bit of litter, it might have a "temperature" of 20 (no big deal).
If a street is covered in graffiti, tangled wires, and massive billboards, its "temperature" might hit 90 (a critical emergency!).

This allows city leaders to look at a digital map and see "hotspots" of visual chaos, telling them exactly where to send the cleaning crews first.

4. The "Master Plan" (The Proposed Framework)

The researchers propose a 5-layer system to turn this into a real-world tool:

The Collectors: Using drones, CCTV cameras, and even citizens' smartphones to gather "eyes" on the street.
The Detectors: AI that spots the mess in real-time.
The Measurers: AI that calculates exactly how much of the view is being "stolen" by the pollution.
The Categorizers: Sorting the mess into groups (e.g., "This is a structural problem" vs. "This is just trash").
The Reporters: Sending an automatic "alert" to city officials with a map and a score.

Why does this matter?

It’s not just about making cities look "pretty" for Instagram. The researchers argue that:

Mental Health: Less visual chaos means less stress and "brain fog" for people living in the city.
Public Health: Messy trash and clogged drains aren't just ugly; they are breeding grounds for mosquitoes and diseases (like Dengue).

In short: The paper is proposing a way to give cities a "digital nervous system" that can sense, measure, and report visual mess, helping urban planners create cleaner, calmer, and healthier places to live.

Technical Summary: A Scoping Review of Deep Learning for Urban Visual Pollution and Proposal of a Real-Time Monitoring Framework

1. Problem Statement

Urban Visual Pollution (UVP)—defined as the presence of disorganized, intrusive, or aesthetically disruptive elements such as excessive signage, misplaced waste, chaotic wiring, and structural defects—is a growing global concern. Unlike physical pollution, UVP primarily impacts the psychological well-being of citizens, contributing to cognitive stress and reduced quality of life.

Current management of UVP relies on manual, qualitative, and subjective surveys, which are unscalable, inconsistent, and time-consuming. While computer vision offers a path toward automation, existing research is fragmented: studies are often domain-specific (focusing on a single pollutant), geographically skewed, and lack a unified taxonomy or a standardized method to quantify the severity of pollution.

2. Methodology

The authors employed a dual-phase approach:

Phase I: Scoping Review: Following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines, the researchers systematically searched seven major academic databases (including Scopus, IEEE Xplore, and Web of Science). From an initial pool of 3,439 articles, 26 high-quality, peer-reviewed studies focusing on deep learning-based UVP detection were selected for analysis.
Phase II: Framework Proposal: Based on the gaps identified in the review, the authors designed a multi-layered, end-to-end conceptual framework for real-time UVP monitoring.

3. Key Contributions

The paper provides three primary contributions to the field:

Comprehensive Literature Mapping: A systematic categorization of existing deep learning architectures (CNNs, YOLO variants, Transformers), datasets, and application-level implementations used in UVP research.
Proposed Unified Taxonomy: To resolve the lack of standardization, the authors propose a hierarchical taxonomy of visual pollutants, categorized into:
- Structural/Road Defects (potholes, dilapidated facades).
- Advertisement/Signage (billboards, illegal posters).
- Utility Cables (overhead wires, telecom towers).
- Litter/Waste (street trash, industrial waste).
- Graffiti/Defacement.
The Real-Time Monitoring Framework: A five-layer architectural design:
1. Data Acquisition: Multi-source input (CCTV, UAVs, smartphones) with metadata.
2. Detection Layer: Utilizing lightweight, real-time models (e.g., YOLOv8) for object localization.
3. Segmentation Layer: Pixel-level delineation to calculate the area ratio of pollutants.
4. Classification Layer: Assigning human-understandable categories and severity levels.
5. Reporting & Indexing Layer: The introduction of a mathematical Visual Pollution Index (VPI).

4. Results and Technical Insights

Model Trends: The review found that YOLO variants (v3–v8) are the most widely adopted due to their optimal balance of inference speed and accuracy, making them suitable for edge deployment. Faster R-CNN and Mask R-CNN are used for higher accuracy but suffer from high latency.
Dataset Scarcity: While datasets like TACO (trash) and various regional datasets (e.g., Saudi Arabia Public Roads, VP-Dhaka) exist, they are often class-imbalanced, geographically limited, or lack standardized COCO-style annotations.
The Visual Pollution Index (VPI): The authors propose a novel formula to quantify urban aesthetics:
$VPI = 100 \times (\alpha D + \beta S + \gamma R) \times W$
Where $D$ is Pollutant Density, $S$ is Severity/Area Ratio, $R$ is Recurrence Frequency, $W$ is Spatial Context Weight (e.g., proximity to hospitals), and $\alpha, \beta, \gamma$ are tunable weights.

5. Significance

This research moves the conversation from "detecting objects" to "managing urban environments." The significance lies in its interdisciplinary utility:

Urban Planning: Provides data-driven heatmaps to help planners prioritize infrastructure repairs and aesthetic interventions.
Public Health: Connects UVP to pandemic control (e.g., identifying unmanaged waste that serves as mosquito breeding grounds for Dengue).
Smart City Integration: Offers a scalable blueprint for integrating AI-driven environmental monitoring into existing municipal management systems, transforming UVP management from a reactive, report-based process into a proactive, real-time paradigm.