https://doi.org/10.5061/dryad.cfxpnvxdc

According to the ratio of 2:1, all samples of the six species were divided into training set and test set. We use the number from 0 to 5 to show the species of G.sarpedon, G.milon, G.doson, G.chironides, G.eurypylus and G.meyeri.

Description of the data and file structure

Butterflies (Lepidoptera: Rhopalocera) are the second largest species of insects, with approximately 17,000 recorded species on earth. The species classification and identification of butterfly have been a significant area of research in the field of entomological taxonomy for a long time. In this work, six butterfly species in the genus of Graphium are selected for experiments, including G.sarpedon, G.milon, G.doson, G.chironides, G.eurypylus and G.meyeri. The total number of samples is 36, and for each species the number is 6. All the specimens were made using standard method including softening, pinning, stretching wings, and drying. The boundary shape, spot distribution, and color of the Graphium samples all have high similarity. Therefore, fast discrimination between some species of the samples is especially difficult even for professional scientists.

Considering the possibility of field-use, we choose a hand-held hyperspectral camera of Specim IQ (Specim Spectral Imaging Ltd.) with the wavelength range of 400 to 1000 nm and a spectral resolution of 7 nm. It’s a line scan camera based on push-broom technology with 204 spectral bands. To minimize light scattering errors, the specimens of butterfly and standard white reference (WR) were placed flat on a black background with low-reflectance. The camera lens was oriented vertically downward, with a distance of 26 cm to butterfly specimen. Clear hyperspectral images of the samples were obtained by adjusting suitable focal length and exposure time, for this work the exposure time is 20 ms. The resulting hyperspectral images, aside from the actual raw image data of the target area, were further processed to yield reflectance spectral data of every pixel. Considering that both spectral ends of the camera have a lower signal-to-noise ratio (SNR), only the spectral range of 420-990 nm was used in this work, including 191 spectral bands.

training_set

wb: wb is the abbreviation of waveband, and its unit is 'nm'. The first row of the datasheet lists the waveband values, corresponding to the spectral wavelengths of each band.

Class label (first column): The first column indicates the class label of each sample. A total of six classes are included in the dataset, encoded as integers from 0 to 5.

Spectral data: Each row (excluding the header) represents one sample. Each column (excluding the first column) represents the reflectance value at a specific waveband. Therefore, each row corresponds to a complete spectral curve of a single sample.

Measurement unit: The spectral values represent reflectance, which is a dimensionless quantity (or normalized reflectance).

validation_set

wb: wb is the abbreviation of waveband, and its unit is 'nm'. The first row of the datasheet lists the waveband values, corresponding to the spectral wavelengths of each band.

Class label (first column): The first column indicates the class label of each sample. A total of six classes are included in the dataset, encoded as integers from 0 to 5.

Spectral data: Each row (excluding the header) represents one sample. Each column (excluding the first column) represents the reflectance value at a specific waveband. Therefore, each row corresponds to a complete spectral curve of a single sample.

Measurement unit: The spectral values represent reflectance, which is a dimensionless quantity (or normalized reflectance).

Key Information Sources

The spectral data were acquired using the Specim IQ hyperspectral camera.