Abstract:With consecutive bands in a wide range of wavelengths, hyperspectral images (HSI) have provided a unique tool for object detection task. However, existing HSI object detection methods have not been fully utilized in real applications, which is mainly resulted by the difference of spatial and spectral resolution between the unlabeled target domain and a labeled source domain, i.e. the domain shift of HSI. In this work, we aim to explore the unsupervised cross-domain object detection of HSI. Our key observation is that the local spatial-spectral characteristics remain invariant across different domains. For solving the problem of domain-shift, we propose a HSI cross-domain object detection method based on spectral-spatial feature alignment, which is the first attempt in the object detection community to the best of our knowledge. Firstly, we develop a spectral-spatial alignment module to extract domain-invariant local spatial-spectral features. Secondly, the spectral autocorrelation module has been designed to solve the domain shift in the spectral domain specifically, which can effectively align HSIs with different spectral resolutions. Besides, we have collected and annotated an HSI dataset for the cross-domain object detection. Our experimental results have proved the effectiveness of HSI cross-domain object detection, which has firstly demonstrated a significant and promising step towards HSI cross-domain object detection in the object detection community.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the domain shift problem in hyperspectral image (HSI) cross - domain object detection. Specifically, the existing hyperspectral image object detection methods are not fully utilized in practical applications, mainly due to the domain shift problem caused by the differences in spatial and spectral resolutions between the unlabeled target domain and the labeled source domain. To meet this challenge, the authors propose an unsupervised cross - domain object detection method based on spectral - spatial feature alignment (SFA), aiming to solve the domain shift problem by extracting and aligning local spectral - spatial features in different domains. The following are the main contributions of this method: 1. **For the first time, introduce the HSI cross - domain object detection framework (SFA) based on spectral - spatial feature alignment** to solve the domain shift problem between the source domain and the target domain. 2. **Propose the spectral - spatial alignment module (SSAM) and the spectral auto - correlation module (SACM)** to solve the domain shift problem in the spectral and spatial dimensions. 3. **Collect and label a data set specifically for HSI cross - domain object detection**. Experimental results show that the proposed SFA method is superior to the existing unsupervised cross - domain object detection methods in performance. ### Core problems of the paper - **Domain shift problem**: The differences in spatial and spectral resolutions between the source domain and the target domain lead to domain shift, making it difficult for the existing HSI object detection methods to be directly applied to practical scenarios. - **Unsupervised cross - domain object detection**: How to effectively transfer the detection model of the source domain to the target domain without labels in the target domain and maintain high detection accuracy. ### Solutions - **Spectral - spatial feature alignment module (SSAM)**: Extract local spectral - spatial features through an auto - encoder (AE) and use a gradient reversal layer (GRL) to align these features. - **Spectral auto - correlation module (SACM)**: Calculate the spectral auto - correlation matrices of the source domain and the target domain and minimize the mean square error (MSE) between them to further align the spectral features. - **Data set construction**: Collect and label a new HSI data set, Longwang Port (LWP), to verify the effectiveness of the proposed method. ### Experimental results Experimental results show that the proposed SFA method has achieved significant performance improvement in cross - domain object detection tasks, especially significantly superior to other existing methods in terms of AP (Average Precision) and AR (Average Recall) metrics. ### Formula summary - **Reconstruction loss**: \[ L_r^s=\|\text{ae}(x_S^i)-x_S^i\|_F^2+\alpha\|e_n^3(x_S^i)\|_1 \] \[ L_r^t=\|\text{ae}(x_T^i)-x_T^i\|_F^2+\alpha\|e_n^3(x_T^i)\|_1 \] - **Domain classifier loss**: \[ L_d^s =-(1 - \lambda)\frac{1}{\beta}\log[\text{sigmoid}(-\beta D_o^s)] \] \[ L_d^t=-\lambda\frac{1}{\beta}\log[\text{sigmoid}(-\beta D_o^t)] \] - **Spectral auto - correlation module loss**: \[ L_{SACM}=\|(F_T)^T F_T-(F_S)^T F_S\|_F^2 \] Through these technical means, the paper successfully solves the domain shift problem in HSI cross - domain object detection and provides new ideas and methods for future research.

Hyperspectral Image Cross-Domain Object Detection Method based on Spectral-Spatial Feature Alignment

Cross-Scene Hyperspectral Image Classification Based on Graph Alignment and Distribution Alignment

Salient Object Detection In Hyperspectral Imagery

Feature Selection for Cross-Scene Hyperspectral Image Classification Using Cross-Domain ReliefF

Dictionary Learning-Based Feature-Level Domain Adaptation for Cross-Scene Hyperspectral Image Classification

Cross-Scene Feature Selection For Hyperspectral Images Based On Cross-Domain Information Gain

Cross-Scene Hyperspectral Feature Selection Via Hybrid Whale Optimization Algorithm with Simulated Annealing

Cross-domain Object Detection by Local to Global Object-Aware Feature Alignment

Cross-domain attention network for hyperspectral image classification

Object Detection in Hyperspectral Image via Unified Spectral–Spatial Feature Aggregation

Open-Set Cross-Domain Hyperspectral Image Classification Based on Manifold Mapping Alignment

Spatial Alignment for Unsupervised Domain Adaptive Single-Stage Object Detection

Coarse-to-Fine Joint Distribution Alignment for Cross-Domain Hyperspectral Image Classification

Adversarial Domain Alignment With Contrastive Learning for Hyperspectral Image Classification

SAR-CDSS: A Semi-Supervised Cross-Domain Object Detection from Optical to SAR Domain

Enhanced Cross-Domain Dim and Small Infrared Target Detection via Content-Decoupled Feature Alignment

HyUniDA: Breaking Label Set Constraints for Universal Domain Adaptation in Cross-Scene Hyperspectral Image Classification

Mind the Gap: Multilevel Unsupervised Domain Adaptation for Cross-Scene Hyperspectral Image Classification

Cross-Domain Classification of Multisource Remote Sensing Data Using Fractional Fusion and Spatial-Spectral Domain Adaptation

A Semi-Supervised Domain Alignment Transformer for Hyperspectral Images Change Detection

Deep Latent Spectral Representation Learning-Based Hyperspectral Band Selection for Target Detection