Nondestructive detection of yellow peach quality parameters based on 3D-CNN and hyperspectral images

Abstract

In recent years, efficient and accurate detection of fruit quality has become an important means of non-destructive testing of fruit quality. In order to solve the problems of cost, efficiency and accuracy in the non-destructive testing of yellow peach quality. This paper proposes a new method for simultaneous detection of yellow peach hyperspectral multiple quality parameters. This method uses the method of extracting and recombining wavelengths at equal intervals in the full-band spectrum instead of the characteristic wavelength selection method. It can make the hyperspectral image contain all-band spectral information. This method uses 3D-CNN to replace the original regression modelling method, which improves the accuracy of model prediction. This method uses the shared network convolutional layer method to perform multi-task learning on the sugar content and hardness of yellow peaches, and finally realizes the simultaneous detection of multiple quality parameters of yellow peaches.