| 刊名 | Agricultural Biotechnology |
| 作者 | Guannan WANG1, Shanshan TANG1，Na WANG2* |
| 作者单位 | 1. Heilongjiang Bayi Agricultural University, Daqing 163319, China; 2. College of Information and Electrical Engineering, Heilongjiang Bayi Agricultural University, Daqing 163319, China |
| DOI | DOI:10.19759/j.cnki.2164-4993.2025.01.013 |
| 年份 | 2025 |
| 刊期 | 1 |
| 页码 | 57-61 |
| 关键词 | Near infrared spectroscopy technology; Blueberry; Deep learning; Origin identification |
| 摘要 | [Objectives] This study was conducted to realize the rapid and nondestructive identification of blueberry producing areas and protect benefits of high-quality blueberry brands. [Methods] Five types of blueberries from different regions were selected as experimental subjects, and spectral analysis techniques were combined with deep learning. Firstly, standard normal variable transform (SNV) and convolutional smoothing (SG) were used to deal with scattering noise and other issues in original spectral data. Secondly, due to a large amount of redundant information and high correlation between adjacent wavelengths in the collected spectra, continuous projection algorithm (SPA) and partial least squares regression (PLS) were combined for screening of features with RMSE as the indicator, and 40 feature variables were obtained. Finally, a convolutional network model CNN-SE integrating a Squeeze and Excitation (SE) attention mechanism module was constructed and compared with convolutional neural network (CNN), support vector machine (SVM), and BP neural network. [Results] The CNN-SE model had the best effect, with the accuracy and precision of the test set reaching 95% and 94.56%, respectively, and the recall and F1 score reaching 93.94% and 94.24%, respectively,. [Conclusions] The CNN-SE convolution network model can realize rapid, nondestructive and high-throughout identification of blueberry producing areas. |
