Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AI-enabled Satellite Edge Computing: A Single-Pixel Feature based Shallow Classification Model for Hyperspectral Imaging

Li Fang, Tianyu Li, Yanghong Lin, Shudong Zhou, Wei Yao

TL;DR

The paper tackles the challenge of real-time hyperspectral image classification onboard satellites under bandwidth and resource constraints. It presents a lightweight, non-deep-learning framework that uses a two-stage pixel-wise label propagation with anchor graphs based on single-pixel spectral features, complemented by a rank-constrained sparse graph clustering to determine anchor labels. The approach achieves strong accuracy on Indian Pines, Salinas, and Pavia University datasets, demonstrates robustness to sensor noise and limited labeled data, and runs in linear time suitable for onboard deployment. This work offers practical significance in enabling autonomous decision-making on satellites with limited compute and downlink bandwidth, and it provides insights into parameter choices and efficiency trade-offs, with code to be released publicly.

Abstract

As the important component of the Earth observation system, hyperspectral imaging satellites provide high-fidelity and enriched information for the formulation of related policies due to the powerful spectral measurement capabilities. However, the transmission speed of the satellite downlink has become a major bottleneck in certain applications, such as disaster monitoring and emergency mapping, which demand a fast response ability. We propose an efficient AI-enabled Satellite Edge Computing paradigm for hyperspectral image classification, facilitating the satellites to attain autonomous decision-making. To accommodate the resource constraints of satellite platforms, the proposed method adopts a lightweight, non-deep learning framework integrated with a few-shot learning strategy. Moreover, onboard processing on satellites could be faced with sensor failure and scan pattern errors, which result in degraded image quality with bad/misaligned pixels and mixed noise. To address these challenges, we develop a novel two-stage pixel-wise label propagation scheme that utilizes only intrinsic spectral features at the single pixel level without the necessity to consider spatial structural information as requested by deep neural networks. In the first stage, initial pixel labels are obtained by propagating selected anchor labels through the constructed anchor-pixel affinity matrix. Subsequently, a top-k pruned sparse graph is generated by directly computing pixel-level similarities. In the second stage, a closed-form solution derived from the sparse graph is employed to replace iterative computations. Furthermore, we developed a rank constraint-based graph clustering algorithm to determine the anchor labels.

AI-enabled Satellite Edge Computing: A Single-Pixel Feature based Shallow Classification Model for Hyperspectral Imaging

TL;DR

The paper tackles the challenge of real-time hyperspectral image classification onboard satellites under bandwidth and resource constraints. It presents a lightweight, non-deep-learning framework that uses a two-stage pixel-wise label propagation with anchor graphs based on single-pixel spectral features, complemented by a rank-constrained sparse graph clustering to determine anchor labels. The approach achieves strong accuracy on Indian Pines, Salinas, and Pavia University datasets, demonstrates robustness to sensor noise and limited labeled data, and runs in linear time suitable for onboard deployment. This work offers practical significance in enabling autonomous decision-making on satellites with limited compute and downlink bandwidth, and it provides insights into parameter choices and efficiency trade-offs, with code to be released publicly.

Abstract

As the important component of the Earth observation system, hyperspectral imaging satellites provide high-fidelity and enriched information for the formulation of related policies due to the powerful spectral measurement capabilities. However, the transmission speed of the satellite downlink has become a major bottleneck in certain applications, such as disaster monitoring and emergency mapping, which demand a fast response ability. We propose an efficient AI-enabled Satellite Edge Computing paradigm for hyperspectral image classification, facilitating the satellites to attain autonomous decision-making. To accommodate the resource constraints of satellite platforms, the proposed method adopts a lightweight, non-deep learning framework integrated with a few-shot learning strategy. Moreover, onboard processing on satellites could be faced with sensor failure and scan pattern errors, which result in degraded image quality with bad/misaligned pixels and mixed noise. To address these challenges, we develop a novel two-stage pixel-wise label propagation scheme that utilizes only intrinsic spectral features at the single pixel level without the necessity to consider spatial structural information as requested by deep neural networks. In the first stage, initial pixel labels are obtained by propagating selected anchor labels through the constructed anchor-pixel affinity matrix. Subsequently, a top-k pruned sparse graph is generated by directly computing pixel-level similarities. In the second stage, a closed-form solution derived from the sparse graph is employed to replace iterative computations. Furthermore, we developed a rank constraint-based graph clustering algorithm to determine the anchor labels.
Paper Structure (22 sections, 1 theorem, 22 equations, 9 figures, 8 tables, 2 algorithms)

This paper contains 22 sections, 1 theorem, 22 equations, 9 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Result
  3. Benchmark
  4. Indian Pines
  5. Salinas
  6. Pavia University
  7. Simulation of onboard processing under spaceborne conditions
  8. Semi-supervised Classification Results
  9. Impact of the Number of Labeled Samples
  10. Unsupervised Classification Results
  11. Simulation Study in terms of Sensor Noise
  12. Ablation Study
  13. Parameter Sensitivity
  14. Onboard Computational Efficiency Analysis
  15. Discussion
  16. ...and 7 more sections

Key Result

Theorem 1

The multiplicity $c$ of the eigenvalue zero of the Laplacian matrix $L_{A}$ is equal to the number of connected components in the graph associated with $A$.

Figures (9)

  • Figure 1: A comparison of the classification performance of the proposed method and previous method under the interference of strip noise. (a) The falsecolor image of Pavia University with deadline noise. (b) The classification map with SGLsellars2020superpixel. (c) The classification map using our method.
  • Figure 2: Ground-truth map and classification maps on the Indian Pines dataset using different training samples per class. (a) ground truth. (b) $3$ training samples per class. (c) $5$ training samples per class. (d) $10$ training samples per class. (e) $30$ training samples per class.
  • Figure 3: Ground-truth map and classification maps on the Salinas dataset using different training samples per class. (a) ground truth. (b) $3$ training samples per class. (c) $5$ training samples per class. (d) $10$ training samples per class. (e) $30$ training samples per class.
  • Figure 4: Ground-truth map and classification maps on the Pavia University dataset using different training samples per class. (a) ground truth. (b) $3$ training samples per class. (c) $5$ training samples per class. (d) $10$ training samples per class. (e) $30$ training samples per class.
  • Figure 5: Sensitivity analysis on the value of $\sigma^{2}$ on the Indian Pines, Salinas, and Pavia University datasets. (a) Indian Pines. (b) Salinas. (c) Pavia University.
  • ...and 4 more figures

Theorems & Definitions (1)

  • Theorem 1