Classification and Recognition of Building Appearance Based on Optimized Gradient-Boosted Decision Tree Algorithm

Mengting Hu, Ling‐Xiang Guo, Jing Liu, Yuxuan SongSensors 2023 · DOI

TL;DR

This paper introduces an optimized Gradient-Boosted Decision Tree (GBDT) algorithm for accurately classifying urban land use types by integrating business point-of-interest (POI) data with architectural form features. By developing a novel "form database" and employing supervised learning with k-fold cross-validation, the method achieves high recognition accuracy, particularly for complex building types.

What Problem Does This Paper Address?

Urban planning and development face a significant challenge in accurately identifying and classifying diverse land use types, especially in densely populated urban areas with complex and mixed land uses. Traditional methods, like manual surveys, are time-consuming and resource-intensive, while existing automated approaches using only POI data often struggle with mixed-use plots and achieving the precision required by national land classification standards. This paper aims to bridge this gap by developing a more robust and accurate automated system for urban land use identification.

Methods

The core of this research lies in an optimized Gradient-Boosted Decision Tree (GBDT) algorithm. The authors propose a multi-source data fusion approach, combining:

  • Business-type POI data: This includes geographic coordinates, industry classifications, and names of businesses.
  • Architectural form data: This encompasses morphological indicators like building height, base area, and plot ratio.

Key methodological steps include:

  • Data Preprocessing and Database Construction: POI data is calibrated for spatial accuracy, and a "form database" is innovatively established to store architectural features.
  • Weighted Transformation: The TF-IDF algorithm is used to reweight business data based on its frequency and importance, addressing the issue of magnitude differences between business types.
  • Supervised Classification Learning: A GBDT model is trained using the combined weighted business and architectural form data.
  • Parameter Optimization: The GBDT model's parameters (e.g., number of nodes, maximum depth, learning rate) are iteratively adjusted based on a validation set to achieve optimal performance.
  • K-fold Cross-Validation: This technique is employed to prevent overfitting and ensure the model's generalization ability.
  • Intelligent Rules: Specific rules based on indicator thresholds, architectural layout patterns, and special architectural forms are integrated to further refine classification accuracy, especially for challenging cases.
  • Multi-Type City Sample Learning: The system is trained on diverse city samples to create model clusters adaptable to different city sizes and types.

Datasets

The study utilizes:

  • Business-type POI data: Obtained from network open-source data.
  • Architectural space vector data: Including urban roads, blocks, and building vector data (e.g., height information).
  • Land use data: Divided by standard subcategories, used as machine learning labels.

The primary case study city for training and evaluation is Shenzhen, China, using data from 2018. The paper also mentions extending the method to other cities like Nantong, Changzhou, and Weihai.

Key Insights and Ideas

  • Synergy of POI and Form Data: The most significant insight is that combining business type information with architectural form features dramatically improves land use classification accuracy, overcoming limitations of using either data source alone.
  • The "Form Database" Innovation: Creating a dedicated database for architectural form indicators is a novel contribution that allows for a more structured and comprehensive integration of geometric and spatial building characteristics.
  • TF-IDF for Business Data Normalization: Applying TF-IDF to POI data effectively addresses the imbalance in the number of businesses across different categories, preventing smaller categories from being overshadowed.
  • Intelligent Rules for Edge Cases: The development of specific indicator, layout, and special form rules is crucial for handling complex or ambiguous cases that standard algorithms might misclassify.
  • City-Size Adaptive Models: The concept of training city-size-specific model clusters allows for more tailored and accurate recognition across different urban scales.

Key Results

  • The proposed optimized GBDT algorithm achieves high accuracy in building recognition.
  • Specifically, for R, S, and U-class buildings, the overall accuracy rate of recognition reaches over 94%.
  • For Class A and Class B land, recognition is accurate for small categories, exceeding 80%.
  • For R, M, W, S, and U categories, accuracy for the middle class exceeds 85%.
  • The method outperforms previous approaches in recognizing R, S, and U land types.

Limitations and Future Work

The authors acknowledge several limitations:

  • Residential Environment and Building Quality: The current model cannot fully capture nuances of residential environments and building quality solely through business type and form data, making it difficult to distinguish certain R-type land uses (e.g., low-rise commercial housing vs. temporary housing).
  • Team Knowledge Level: Limitations in team expertise and the scope of business/form data considered.

Future work suggestions include:

  • Incorporating Architectural Textures: To better distinguish land types with similar architectural forms but different textures (e.g., homogenized vs. irregular layouts).
  • Leveraging Multi-Source Big Data and Trusted AI: Exploring deeper integration of diverse data sources and advancing AI technologies for urban research and planning applications.

Summary updated

Abstract

There are high concentrations of urban spaces and increasingly complex land use types. Providing an efficient and scientific identification of building types has become a major challenge in urban architectural planning. This study used an optimized gradient-boosted decision tree algorithm to enhance a decision tree model for building classification. Through supervised classification learning, machine learning training was conducted using a business-type weighted database. We innovatively established a form database to store input items. During parameter optimization, parameters such as the number of nodes, maximum depth, and learning rate were gradually adjusted based on the performance of the verification set to achieve optimal performance on the verification set under the same conditions. Simultaneously, a k-fold cross-validation method was used to avoid overfitting. The model clusters trained in the machine learning training corresponded to various city sizes. By setting the parameters to determine the size of the area of land for a target city, the corresponding classification model could be invoked. The experimental results show that this algorithm has high accuracy in building recognition. Especially in R, S, and U-class buildings, the overall accuracy rate of recognition reaches over 94%.
BibTeX
@inproceedings{hu2023classification,
  author = {Mengting Hu and Ling‐Xiang Guo and Jing Liu and Yuxuan Song},
  title = {{Classification and Recognition of Building Appearance Based on Optimized Gradient-Boosted Decision Tree Algorithm}},
  booktitle = {Sensors},
  year = {2023},
  doi = {10.3390/s23115353},
}
Figure