•  
  •  
 

Turkish Journal of Agriculture and Forestry

Abstract

Spectral analysis methods can nondestructively assess the quality of agricultural products but are limited by imaging speed and equipment costs. Therefore, they have not yet been fully integrated into engineering practices. This paper presents JujubeNet, a two-stage hyperspectral reconstruction model for detecting moisture content in jujubes. The first stage utilizes a generative pretrained Pix2Pix generative adversarial network (GAN) to generate RGB images of jujubes that mimic real camera styles and are pixel-to-pixel aligned with hyperspectral images. The second stage uses a transformer-based spectral reconstruction model to achieve hyperspectral reconstruction. Additionally, we developed a moisture content detection model based on hyperspectral images, integrating it as a downstream task of JujubeNet through cascaded training with loss functions and module attributes. Experiments showed that RGB images generated by the Pix2Pix GAN improved the accuracy of various spectral reconstruction models, with JujubeNet achieving mean relative absolute error of 0.1501, outperforming other models. In downstream moisture-content detection, JujubeNet achieved a root mean square error of 0.0064—superior to other networks. In conclusion, JujubeNet delivers satisfactory results in both hyperspectral images generation and moisture content detection. The model offers a nondestructive, efficient, and low-cost solution for crop quality monitoring, thus promoting agricultural sustainability.

Author ORCID Identifier

YALEI XU: 0000-0002-8609-0558

YANG LI: 0000-0002-4268-4004

JING NIE: 0000-0002-3763-9559

JINGBIN LI: 0000-0003-4264-7024

DOI

10.55730/1300-011X.3326

Keywords

Jujube, image generation, hyperspectral reconstruction, moisture content detection

First Page

1095

Last Page

1105

Publisher

The Scientific and Technological Research Council of Türkiye (TÜBİTAK)

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download

Share

COinS