Research map / Image generation / Image restoration
Hyper-Spectral Image Generation from Frequency Spectrums
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
A PhD thesis proposal on automatically constructing hyper-spectral images from DAS/radargram frequency spectrums by learning frequency-band filters jointly with an object detector, using greedy and simulated annealing search.
Problem
Frequency spectrums from sensing systems (DAS, radargrams) are too wide to map one frequency per channel, so frequency bands must be chosen by hand; this paper asks how hyper-spectral images can be automatically built from spectra for downstream object detection.
Approach
Learn frequency-band filters jointly with the object detector, which acts as both measure and selector: greedy selection and greedy set search prune/split bands by detector performance, and a simulated annealing variant changes band configurations during training to cut training time.
Key results
- Greedy selection and greedy set search reportedly outperform the current state-of-the-art and expert-knowledge bands but require several days of training.
- Simulated annealing based filter learning trains in several hours and performs better than baselines but does not outperform the greedy methods.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use greedy feature forward selection to choose and refine frequency bands | DAS hyper-spectral image generation with an object detector · DAS object detection | detection performance | Helped |
| Use greedy set search for feature backward selection | DAS hyper-spectral image generation with an object detector · DAS object detection | detection performance | Helped |
| Use simulated annealing to change filter configurations during detector training | DAS hyper-spectral image generation with an object detector · DAS object detection | detection performance | Helped |
| Replace greedy selection or greedy set search with simulated annealing | DAS hyper-spectral image generation with an object detector · DAS object detection | detection performance | Hurt |
Limitations
- Quantitative results are not reported in the extract
- Evaluated mainly on proprietary DAS data from an industry partner
- Greedy methods require days of training
Experimental setup
Datasets: DAS system frequency spectrums from Future Fibre Technologies
Baselines: expert-knowledge frequency bands baseline methods (unspecified)
Builds on: Fast Fourier Transform
Closest papers by meaning
- Acoustic-to-Hyper-Spectral: Hyper-Spectral Image Construction from Frequency Spectrums Through Simulated Annealing (Student Abstract) (AAAI 2025 · similarity 0.90)
- FRN: Fractal-Based Recursive Spectral Reconstruction Network (NeurIPS 2025 · similarity 0.82)
- Any-Resolution AI-Generated Image Detection by Spectral Learning (CVPR 2025 · similarity 0.81)
- Hyperspectral Image Reconstruction via Combinatorial Embedding of Cross-Channel Spatio-Spectral Clues (AAAI 2024 · similarity 0.80)
- SPECAT: SPatial-spEctral Cumulative-Attention Transformer for High-Resolution Hyperspectral Image Reconstruction (CVPR 2024 · similarity 0.79)
- Frequency Dynamic Convolution for Dense Image Prediction (CVPR 2025 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.