Dive into Aerial Remote Sensing Underwater Depth Estimation with Hyperspectral Imagery

Authors: Jiahao Qi, Xingyue Liu, Chen Chen, Dehui Zhu, Kangcheng Bin, Ping Zhong

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on the constructed benchmark dataset validate the potential of HUDE and the effectiveness of HUDEMamba. ... The quantitative results of this evaluation are detailed in Table 1, with corresponding visual results shown in Figure 6. In quantitative comparison, HUDEMamba demonstrates superior performance compared to the SOTA methods, achieving higher accuracy metrics while yielding lower values in certain error metrics.
Researcher Affiliation Academia National Key Laboratory of Science and Technology on Automatic Target Recognition National University of Defense Technology, China EMAIL
Pseudocode No The paper describes methods and processes (e.g., HSI Unmixing and Abundance Ranking, Context-aware Representation Learning) but does not include any explicitly labeled pseudocode or algorithm blocks with structured steps.
Open Source Code No The paper provides a link to a dataset: "Datasets https://github.com/qjh1996/ATR-HUDE". However, it does not explicitly state that the source code for the methodology described in the paper is available at this link or elsewhere.
Open Datasets Yes For the benchmark dataset, we construct a real-world hyperspectral UDE (HUDE) dataset ATRHUDE, comprising approximately 500 synchronized hyperspectral and Li DAR data pairs collected from diverse coastal scenes and flight altitudes. ... Datasets https://github.com/qjh1996/ATR-HUDE
Dataset Splits Yes We segmented the processed images into non-overlapping 100 100 blocks, yielding 500 image pairs. We utilized 400 pairs for training and 100 pairs for testing. Image pairs for different tasks were proportionally sampled from diverse scenes and altitudes to minimize data distribution bias.
Hardware Specification Yes HUDEMamba is implemented on seven NVIDIA RTX6000 Ada 48GB GPUs running Ubuntu 22.04.
Software Dependencies No The paper mentions 'Ubuntu 22.04' as the operating system and 'Adam optimizer' as the optimization method, but does not specify version numbers for other key software components or libraries (e.g., Python, PyTorch, TensorFlow).
Experiment Setup Yes For the physics-inspired image translation stage, we utilize VCA (Rasti et al. 2024) as the hyperspectral unmixing (HU) method, setting the hyperparameter N in Equation (2) to 5 for all coastal scenes. ... Additionally, the full depth range [dmin, dmax] in Equation (9) is set to [0.5, 3] meters for all compared methods based on the proposed dataset. HUDEMamba is implemented on seven NVIDIA RTX6000 Ada 48GB GPUs running Ubuntu 22.04. We used the Adam optimizer with a batch size of 8. The initial learning rate of 1 10 4 was gradually reduced to 1 10 6 following a cosine schedule over 300 epochs.