Hyperspectral Image Dataset for Individual Penguin Identification

TL;DR

This work addresses non-invasive identification of individual penguins by leveraging hyperspectral imaging at the single-pixel level. A simple 5-layer MLP, coupled with a denoising step, classifies penguin individuals from the spectral vector of a single HS pixel, and is evaluated against RGB and PCA-reduced HS inputs. The authors introduce a novel HS penguin dataset (990 images, 27 penguins) with pixel-level IDs and bounding boxes, and demonstrate that full HS data yields substantially higher identification accuracy (82.06% OA) than RGB (27.16%) or PCA (51.03%). The results, along with spectral signature analyses across time intervals, suggest HS-based pixel-wise identification is a viable tool for non-invasive wildlife monitoring and can be extended to other species.

Abstract

Remote individual animal identification is important for food safety, sport, and animal conservation. Numerous existing remote individual animal identification studies have focused on RGB images. In this paper, we tackle individual penguin identification using hyperspectral (HS) images. To the best of our knowledge, it is the first work to analyze spectral differences between penguin individuals using an HS camera. We have constructed a novel penguin HS image dataset, including 990 hyperspectral images of 27 penguins. We experimentally demonstrate that the spectral information of HS image pixels can be used for individual penguin identification. The experimental results show the effectiveness of using HS images for individual penguin identification. The dataset and source code are available here: https://033labcodes.github.io/igrass24_penguin/

Figures (9)

• Figure 1: Main idea. (a) Pixels of the target penguins are selected from the captured hyperspectral (HS) images. (b) Pixels selected from the HS images retain information on the wavelength of reflected light. (c) A simple 5-layer MLP is used to predict individual labels of penguins. (d) Individual identification was performed using a single pixel.
• Figure 2: Shooting conditions of the HS images in the dataset.
• Figure 3: Photographic scenery of the HS images in the dataset.
• Figure 4: An example from the dataset: An RGB image created from the HS image.
• Figure 5: Selection of input pixels with an application in mind and visualization of predicted labels. Pixels of the target with individual label 5 is selected, and the output results are colored differently corresponding to each label.