In-line monitoring astaxanthin in krill meal using computer vision

文献类型： 外文期刊

第一作者： Zhang, Quantong

作者： Zhang, Quantong;Yang, Liu;Guo, Quanyou;Zheng, Yao;Zhang, Quantong;Yang, Liu;Guo, Quanyou;Zheng, Yao;Zhang, Quantong;Guo, Quanyou;Zheng, Yao;Zhang, Quantong;Wang, Xin

作者机构：

关键词： Krill meal; Astaxanthin content; Shallow convolutional neural networks; Computer vision; Otsu algorithm

期刊名称：JOURNAL OF FOOD COMPOSITION AND ANALYSIS （ 影响因子：4.6； 五年影响因子：4.6 ）

ISSN： 0889-1575

年卷期： 2025 年 146 卷

页码：

收录情况： SCI

摘要： This study aimed to investigate a method for in-line determination of astaxanthin content in Antarctic krill meal using computer vision and a shallow convolutional neural network (SCNN). A database containing 13,995 images and corresponding labels was established. The images were automatically captured and preprocessed using a threshold segmentation algorithm, and the astaxanthin content labels were determined using liquid chromatography. An SCNN model was developed to predict the astaxanthin content of krill meal based on color features. Based on the developed model, a multithreaded program was proposed for in-line quantification of astaxanthin, integrating functions such as automatic image acquisition, preprocessing, model deployment, prediction, and result display. The model demonstrated strong robustness for images captured at varying conveyor speeds and region of interest (ROI) sizes. At a conveyor speed of 2.5 m/min, for an ROI of 2242 pixels, the model achieved R2, MAE, and RMSE values of 0.934, 2.42, and 3.16, respectively. Evaluation of the in-line systems revealed an absolute error distribution and relative error distribution ranging from-2.40 to 3.73 mg/kg and 1.31-8.60 %, respectively, compared to the observed values. Thus, the in-line system enabled automatic, rapid, nondestructive, and accurate quantification of astaxanthin in krill meal.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序