Discrimination of the fruits of Amomum tsao-ko according to geographical origin by 2DCOS image with RGB and Resnet image analysis techniques

Amomum tsao-ko Crevost et Lemaire is a well-known dietary spice in the world for its unique flavor and medicinal effects. Nevertheless, the geographical origin of A. tsao-ko fruits plays a key role in affecting their aroma of cooking food and medicinal effects. This study attempted to investigate the prospects of using two dimensional correlation spectra (2DCOS) and image analysis methods for tracing the origins of A. tsao-ko fruits. To this goal, the near infrared (NIR) spectra of 439 A. tsao-ko fruits collected from 6 regions were obtained and converted into synchronous and asynchronous 2DCOS images. On this basis, two image analysis methods, including Red-Green-Blue (RGB) image analysis and residual convolutional neural network (Resnet) analysis, were applied for authenticating the geographical origin. The results of two image classification models indicated that synchronous 2DCOS images were more suitable to discriminate the geographical origins of A. tsao-ko fruits than asynchronous 2DCOS images. Furthermore, the comparison of categorized result between two image classification models suggested that Resnet model could not only automatically extract features from raw data but also provide more better discriminate model. Therefore, synchronous 2DCOS images combined Resnet analysis could be used as a reliable method for quality control of spice and herb.