MOB-GCN: A Novel Multiscale Object-Based Graph Neural Network for Hyperspectral Image Classification

TL;DR

This work addresses hyperspectral image classification under limited labeled data by integrating multiscale object-based image analysis with graph neural networks. It introduces MOB-GCN, which leverages Felzenszwalb superpixel segmentation and a multiresolution graph learning framework (MGN) to fuse features across 4–6 optimal segmentation scales, enabling robust semi-supervised classification. The method demonstrates superior Overall Accuracy compared with a single-scale GCN across six benchmark HSIs, with notable gains on small training sets and improved noise robustness. Practically, MOB-GCN combines accuracy and computational efficiency by using a reduced, multiscale graph representation, making it well-suited for remote sensing tasks where labeled data are scarce.

Abstract

This paper introduces a novel multiscale object-based graph neural network called MOB-GCN for hyperspectral image (HSI) classification. The central aim of this study is to enhance feature extraction and classification performance by utilizing multiscale object-based image analysis (OBIA). Traditional pixel-based methods often suffer from low accuracy and speckle noise, while single-scale OBIA approaches may overlook crucial information of image objects at different levels of detail. MOB-GCN addresses this issue by extracting and integrating features from multiple segmentation scales to improve classification results using the Multiresolution Graph Network (MGN) architecture that can model fine-grained and global spatial patterns. By constructing a dynamic multiscale graph hierarchy, MOB-GCN offers a more comprehensive understanding of the intricate details and global context of HSIs. Experimental results demonstrate that MOB-GCN consistently outperforms single-scale graph convolutional networks (GCNs) in terms of classification accuracy, computational efficiency, and noise reduction, particularly when labeled data is limited. The implementation of MOB-GCN is publicly available at https://github.com/HySonLab/MultiscaleHSI

Figures (19)

• Figure 1: The proposed framework follows a structured pipeline for hyperspectral image (HSI) classification. First, the HSI is read and undergoes dimensionality reduction before applying superpixel segmentation. Features are then extracted from each superpixel and, along with the initial labeling, are used to construct a superpixel-based graph. This graph is first processed by a bottom encoder before undergoing recursive pooling and encoding at multiple resolutions. The latent representations from all resolutions, including the bottom encoding, are concatenated and passed into a final classifier. The predicted labels are then mapped back to the superpixel regions, producing the final classification of the HSI.
• Figure 2: The Salinas HSI segmented using Felzenszwalb segmentation algorithm. felzenszwalb2004 The first figure shows a false-colored RGB image and other 3 shows the image segmented using a minimum size of 50, 100, 200 pixels respectively.
• Figure 3: Graph construction visualization for the INDIAN dataset, with node placement based on TSNE embeddings of node features and node labels from GCN inference.
• Figure 4: Comparison of classification maps for different datasets on 5% sample data. Each subfigure shows ground truth labels and classification maps from GCN, MOB-GCN, unoptimized, and optimized models.
• Figure 5: Inertia and Superpixel standard deviation for K-means clustering assessment on the SALINAS dataset.

Theorems & Definitions (2)

• Definition 1: Semi-Supervised Classification Task
• Definition 2: Superpixel Segmentation