太赫兹光谱转图像及ResNet的陈皮年份鉴别

ChenpiAge Identification Based on Terahertz Spectral Imaging and ResNet

中国陈皮(学名柑橘皮,PCR),它们具有高药用价值且随时间增长而增值,但是由于各个年份的陈皮的形态类似,难以区分。因此本研究提出使用太赫兹技术结合深度学习方法对陈皮样品进行贮藏年份的鉴别。利用图像特征转换方法,包括Gramian角场(GAF)、Markov过渡场(MTF)和递归图(RP),将光谱数据成功转化为RGB三维图像数据,构建了ResNet模型,然后将ResNet模型与1D-CNN模型和CNN-LSTM模型进行对比。结果显示,ResNet模型表现最优,准确率达到了0.8035。总之,将太赫兹光谱数据转换为图像数据结合深度学习模型可以有效区分不同贮藏年份的陈皮,为陈皮的年份检测提供了一个有效的方法。

Chinese Chenpi, known scientifically as pericpium citri reticulatae (PCR), possess significant medicinal value and appreciate in worth over time. However, the morphological similarities among Chenpi from different years make them difficult to distinguish. Hence, this study proposes the utilization of terahertz technology in conjunction with deep learning methods for identifying the storage years of mandarin peel samples. Through the application of image feature transformation techniques, such as Gramian Angular Field (GAF), Markov Transition Field (MTF), and Recurrence Plot (RP), the spectral data is successfully transformed into RGB three-dimensional image data. Subsequently, a ResNet model is constructed and compared to 1D-CNN and CNN-LSTM models. The results indicate that the ResNet model outperforms, achieving an accuracy rate of 0.8035. In conclusion, converting terahertz spectral data into image data, in combination with deep learning models, offers an effective means of distinguishing Chenpi of different storage years, thereby providing an efficient approach to mandarin Chenpi year detection.