mepleleo/DA_peanut

DA_peanut

paper: Identification of Moldy Peanuts under Different Varieties and Moisture Content Using Hyperspectral Imaging and Data Augmentation Technologies

classification model:

KNN/SVM/MobileViT-xs

data augmentation methods:

1. pixel classifier(KNN/SVM)
   Erasing
   Noise
   TSW
   DSM(proposed method)

2. image classifier(MobileViT-xs)
   Erasing
   Noise
   Rotation
   DSM(proposed method)

Method principle

1. generated spectra = original spectra + λ × spectral difference
2. generated hyperspectral image = rotate(original image + λ × spectral difference)

λ is a parameter used to adjust the offset of the original data.
For spectral difference:
The data of this class was divided into two parts with the median as the boundary, and the average spectrum of each part was calculated for generating the spectral difference of two parts.