Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantum Feature-Empowered Deep Classification for Fast Mangrove Mapping

Chia-Hsiang Lin, Po-Wei Tang, Alfredo R. Huete

TL;DR

This work tackles mangrove mapping from Sentinel-2 multispectral imagery by introducing QEDNet, a lightweight dual-branch network that fuses affine CNN features with entangled unitary QNN features. The CNN branch is paired with a QNN branch that employs spatial and spectral quantum encoders and a quantum fusion module, with outputs combined as $Y = \mathrm{Sigmoid}(f_{CNN}(X) + f_{QNN}(X))$. Ablation studies show the quantum features provide genuinely new information that improves classification beyond CNN alone, yielding state-of-the-art results across multiple testing regions and regimes while maintaining computational efficiency. The approach offers threshold-free, rapid mangrove mapping with potential for broader remote-sensing applications via fusion of unitary quantum features and conventional CNN features.

Abstract

A mangrove mapping (MM) algorithm is an essential classification tool for environmental monitoring. The recent literature shows that compared with other index-based MM methods that treat pixels as spatially independent, convolutional neural networks (CNNs) are crucial for leveraging spatial continuity information, leading to improved classification performance. In this work, we go a step further to show that quantum features provide radically new information for CNN to further upgrade the classification results. Simply speaking, CNN computes affine-mapping features, while quantum neural network (QNN) offers unitary-computing features, thereby offering a fresh perspective in the final decision-making (classification). To address the challenging MM problem, we design an entangled spatial-spectral quantum feature extraction module. Notably, to ensure that the quantum features contribute genuinely novel information (unaffected by traditional CNN features), we design a separate network track consisting solely of quantum neurons with built-in interpretability. The extracted pure quantum information is then fused with traditional feature information to jointly make the final decision. The proposed quantum-empowered deep network (QEDNet) is very lightweight, so the improvement does come from the cooperation between CNN and QNN (rather than parameter augmentation). Extensive experiments will be conducted to demonstrate the superiority of QEDNet.

Quantum Feature-Empowered Deep Classification for Fast Mangrove Mapping

TL;DR

This work tackles mangrove mapping from Sentinel-2 multispectral imagery by introducing QEDNet, a lightweight dual-branch network that fuses affine CNN features with entangled unitary QNN features. The CNN branch is paired with a QNN branch that employs spatial and spectral quantum encoders and a quantum fusion module, with outputs combined as . Ablation studies show the quantum features provide genuinely new information that improves classification beyond CNN alone, yielding state-of-the-art results across multiple testing regions and regimes while maintaining computational efficiency. The approach offers threshold-free, rapid mangrove mapping with potential for broader remote-sensing applications via fusion of unitary quantum features and conventional CNN features.

Abstract

A mangrove mapping (MM) algorithm is an essential classification tool for environmental monitoring. The recent literature shows that compared with other index-based MM methods that treat pixels as spatially independent, convolutional neural networks (CNNs) are crucial for leveraging spatial continuity information, leading to improved classification performance. In this work, we go a step further to show that quantum features provide radically new information for CNN to further upgrade the classification results. Simply speaking, CNN computes affine-mapping features, while quantum neural network (QNN) offers unitary-computing features, thereby offering a fresh perspective in the final decision-making (classification). To address the challenging MM problem, we design an entangled spatial-spectral quantum feature extraction module. Notably, to ensure that the quantum features contribute genuinely novel information (unaffected by traditional CNN features), we design a separate network track consisting solely of quantum neurons with built-in interpretability. The extracted pure quantum information is then fused with traditional feature information to jointly make the final decision. The proposed quantum-empowered deep network (QEDNet) is very lightweight, so the improvement does come from the cooperation between CNN and QNN (rather than parameter augmentation). Extensive experiments will be conducted to demonstrate the superiority of QEDNet.
Paper Structure (10 sections, 1 theorem, 8 equations, 6 figures, 6 tables)

This paper contains 10 sections, 1 theorem, 8 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Dual-Branch QEDNet for Quantum Feature and Traditional Feature Extractions
  4. CNN Branch
  5. QNN Branch
  6. Experimental Results
  7. Experimental Settings
  8. Quantitative and Qualitative Analyses
  9. Ablation Study
  10. Conclusions and Future Works

Key Result

Theorem 1

The trainable quantum neurons deployed in the proposed quantum spectral encoder (cf. Figure fig:qnnencoder) and QFM (cf. Figure fig:fusion) can express any valid quantum unitary operator $U$, with some real-valued trainable network parameters $\{\beta_{k},\epsilon_{k},\eta_{k},\lambda_{k},\rho_{k}\}

Figures (6)

  • Figure 1: Graphical illustration of the proposed quantum-empowered deep network (QEDNet). It incorporates the entangled quantum features into traditional CNN under a parallel structure, so that both QNN and CNN features can equally contribute to the final classification results. Precisely, QNN captures unitary-computing feature due to the nature of quantum neurons, and this is fundamentally different from the affine-mapping CNN feature, thus further providing new information for making better final decision/classification. For the QNN branch, the detailed structure of the quantum spatial-spectral encoder will be presented in Figure \ref{['fig:qnnencoder']}, and the quantum fusion module will be presented in Figure \ref{['fig:fusion']}.
  • Figure 2: Detailed architecture of the proposed QNN-based spatial encoder and spectral encoder, both composed of quantum unitary operators. Specifically, the spatial encoder entangles/compresses spatial information from each $2\times 2$ local patch into a single pixel, where the local patch is illustrated as the yellow $x_i$'s serving as one input batch of the spatial encoder. Also, the spectral encoder entangles/captures quantum characteristics across various channel groups (cf. Figure \ref{['fig:fusion']}) along the spectral dimension, where one channel group is illustrated as the yellow $y_i$'s serving as one input batch of the spectral encoder.
  • Figure 3: Detailed architecture of the proposed QNN-based feature fusion block (FFB). The quantum FFB aims to further integrate the previously encoded spatial-spectral quantum features while capturing the correlation between the neighboring bands. To obtain the global channel relations in the final QNN features, we first set three local groups to capture their neighboring channel correlations using the quantum fusion module (QFM), and then fuse the three representative locally correlated features. The QFM is designed by rotation gates and Ising gates, followed by entanglement mechanisms, where the used gates are defined in Table \ref{['tab:common_qu_gate']}.
  • Figure 4: Qualitative study using Myanmar data. (a) RGB reference. (b) Ground-truth map. Classification maps obtained by (c) NDVI, (d) MMRI, (e) MVI, (f) GC-UNet, (g) DCNN, (h) ME-Net, (i) Capsules-UNet, and (j) the proposed QEDNet.
  • Figure 5: Qualitative study using Thailand data. (a) RGB reference. (b) Ground-truth map. Classification maps obtained by (c) NDVI, (d) MMRI, (e) MVI, (f) GC-UNet, (g) DCNN, (h) ME-Net, (i) Capsules-UNet, and (j) the proposed QEDNet.
  • ...and 1 more figures

Theorems & Definitions (1)

  • Theorem 1