CHNSpec Technology (Zhejiang)Co.,Ltd chnspec@colorspec.cn 86--13732210605
Orange peel has good economic value and medicinal value, but the phenomenon of fake and shoddy in the market is serious. In particular, as an important index to measure the quality of orange peel, the accuracy and efficiency of manual detection methods are low. In this paper, hyperspectral imaging technology combined with deep learning method was used to establish a fast and non-destructive identification method for the aging year of orange peel.
一、Materials and methods
The purchased orange peel samples were divided into 1 year, 5 years, 10 years and 15 years according to the aging years. As shown in Figure 1, 120 orange peel samples were collected for each year, and a total of 480 orange peel samples were collected. The orange peel samples of each year were randomly divided in a ratio of 7:3, in which 84 samples entered the training set and 36 samples entered the test set.
In this paper, a 900-1700nm hyperspectral camera is used, and FS-15, a product of Color Spectrum Technology (Zhejiang) Co., LTD., can be used for related research. Short-wave near-infrared hyperspectral camera, the acquisition speed of the full spectrum up to 200FPS, is widely used in the composition identification, substance identification, machine vision, agricultural product quality, screen detection and other fields.
二、Results and analysis
The spectral curves of orange peel samples in different years are shown in Figure 3. The original spectral curves shown in Figure 3 can obviously find that there are absorption peaks near 1200m and 1450nm. The absorption peak at 1200nm is mainly caused by the spectral absorption of bond pairs, and the absorption peak at 1450nm is mainly caused by the spectral absorption of water. The bands of the NIR spectra of all kinds of samples overlapped closely, the overall trend was close to the same, and the absorption peak was almost in the same position, with no obvious difference. It was difficult to distinguish the four kinds of orange peel samples by naked eye.
三、Spectral pretreatment method
The pretreatment of hyperspectral data of orange peel includes several steps, which are image segmentation, spectrum averaging and spectrum preprocessing. The original average spectrum of orange peel samples in different years and the average spectral curves after SG+D1 pretreatment are shown in Figure 4. It can be seen from FIG. 4(a) and FIG. 4(b) that the SG+D1 combined pretreatment method can effectively eliminate the influence of spectral baseline drift and smooth the spectral curve, thus improving the accuracy of orange peel year identification.
Rapid identification of orange peel year by hyperspectral camera has broad application prospect in Chinese medicine industry. On the one hand, it can help Chinese medicine manufacturers and dealers accurately control the quality and year of orange peel, and avoid economic losses and reputation risks caused by misjudgment of year. On the other hand, in terms of market supervision, relevant departments can use the technology to carry out rapid sampling of orange peel products on the market, crack down on shoddy and other behaviors, and maintain the normal order of the market. In addition, with the continuous improvement and popularization of the technology, it will also provide strong support for the scientific research and quality evaluation of orange peel, and promote the development of orange peel industry in a more standardized, standardized and scientific direction.