Novel approach of association rule mining for tree canopy assessment

Nilkamal More, V. B. Nikam, Biplab BanerjeeIAES 2021 · DOI

Abstract

The evolution of technology and availability of voluminous satellite images are bringing a new scenario in satellite image classification where a performance efficient method for predictive analysis of satellite images for land cover classification needs to be devised. As urban areas are growing at faster rate, special attention needs to be given to solve tree canopy assessment problem. Vegetation indices are calculated from spectral information of satellite images. Hundreds of such vegetation indices are available to detect vegetation from a satellite image. The contribution of this paper is designing an improved Apriori algorithm to select optimal number of vegetation indices for tree canopy assessment. In this research, we propose a novel computational approach that allows the improvement of results. It selects optimal combination of vegetation indices and applies principal component analysis on it. It uses a greedy approach based on Apriori algorithm. This study emphasizes on assessment of tree canopy using GPU-enabled environment for performance-efficient assessment. The results achieved, are comparable to state-of-the-art techniques, with an accuracy of 96%. The research has considered 4 years data for Mumbai city of India. This research is useful for Green India Mission of India to assess tree canopy of urban region.

Introduction

As urban areas expand rapidly, monitoring green cover becomes both more critical and more difficult. Traditional ground surveys are impractical for mega-cities, leading researchers to rely on satellite imagery. However, "tree canopy assessment" in an urban environment is a complex classification problem due to the high density of roads, buildings, and varying shadows.

While hundreds of "Vegetation Indices" (VIs)—mathematical combinations of different light spectra like NDVI—exist to help identify plants, using all of them creates a "curse of dimensionality" that slows down computation and can actually decrease accuracy. This research tackles the challenge of identifying the optimal subset of these indices to maximize classification performance without overwhelming the system.

Methodology

The researchers developed a multi-stage computational pipeline to efficiently process satellite data and identify tree canopies.

Methodology Flowchart
  1. Modified Apriori Algorithm: Traditionally used in data mining to find frequent itemsets in transactions, the authors adapted this greedy approach for feature selection. It treats different vegetation indices as "items" and uses classification accuracy as the threshold to prune ineffective combinations and "join" successful ones into an optimal feature set.
  2. Dimensionality Reduction (PCA): After selecting the best indices and spectral bands, the team applied Principal Component Analysis. This transformed the high-dimensional data into a smaller set of uncorrelated components, with the first three components capturing the vast majority of the variance necessary for classification.
  3. SVM Classification: The final processed data was fed into a Support Vector Machine (SVM) using the LibSVM library with a Radial Basis Function (RBF) kernel to perform binary classification (tree vs. non-tree).
  4. High-Performance Computing: The entire workflow was implemented using Google Earth Engine (GEE), leveraging its GPU/TPU-enabled cloud environment to process voluminous satellite datasets efficiently.

Datasets and Study Area

The primary data source for this study was Sentinel-2 satellite imagery, providing 13 spectral bands with spatial resolutions ranging from 10m to 60m. The study area focused on Mumbai and Navi Mumbai, India (approximately 964 sq. km), covering four years of data from 2016 to 2019. To create the ground truth for the supervised learning model, 2,200 geometrical objects were manually annotated via Google Earth Engine.

Key Insights and Results

The core innovation of this work lies in Feature Synergy. Certain combinations of vegetation indices work better together than others. The Apriori-based selection avoids the brute-force approach of testing every possible combination, significantly reducing computational overhead. Furthermore, the study highlights that classification accuracy is highly dependent on the season. By analyzing three distinct periods (Jan-March, April-June, Oct-Dec), they found that specific windows provide more consistent environmental conditions for accurate canopy detection.

Performance Metrics

  • High Accuracy: The proposed method achieved an overall accuracy ranging from 96.67% to 98.94%, depending on the season and year.
  • Thematic Accuracy: The average thematic accuracy was between 91.10% and 96.49%.
  • Benchmarking: These results represent a significant improvement over state-of-the-art techniques, which typically achieve around 90% accuracy.
  • PCA Efficiency: The researchers found that the first three principal components were sufficient for high-accuracy classification, allowing them to discard the "noise" found in higher-order components.

Conclusion and Future Work

The authors successfully mapped canopy changes over a four-year period, providing a tool that can directly support India's "Green India Mission" (GIM). While the current model only performs binary classification (tree vs. non-tree), future iterations could incorporate advanced deep learning architectures like CNNs to improve spatial feature extraction and potentially differentiate between tree species or health status. Despite challenges like atmospheric interference from smog or extreme weather, this novel framework offers a robust and efficient approach to urban environmental monitoring.

Summary updated

Abstract

The evolution of technology and availability of voluminous satellite images are bringing a new scenario in satellite image classification where a performance efficient method for predictive analysis of satellite images for land cover classification needs to be devised. As urban areas are growing at faster rate, special attention needs to be given to solve tree canopy assessment problem. Vegetation indices are calculated from spectral information of satellite images. Hundreds of such vegetation indices are available to detect vegetation from a satellite image. The contribution of this paper is designing an improved Apriori algorithm to select optimal number of vegetation indices for tree canopy assessment. In this research, we propose a novel computational approach that allows the improvement of results. It selects optimal combination of vegetation indices and applies principal component analysis on it. It uses a greedy approach based on Apriori algorithm. This study emphasizes on assessment of tree canopy using GPU-enabled environment for performance-efficient assessment. The results achieved, are comparable to state-of-the-art techniques, with an accuracy of 96%. The research has considered 4 years data for Mumbai city of India. This research is useful for Green India Mission of India to assess tree canopy of urban region.
BibTeX
@inproceedings{more2021novel,
  author = {Nilkamal More and V. B. Nikam and Biplab Banerjee},
  title = {{Novel approach of association rule mining for tree canopy assessment}},
  booktitle = {IAES},
  year = {2021},
  doi = {10.11591/ijai.v10.i3.pp771-779},
}
Figure