Abstract Carrot has high nutritional value and health-promoting effects and is popular among consumers. Real-time quality detection of dried carrot slices allows producers to adjust the process parameters of the drying device in time, thereby ensuring the final product quality and realizing energy conservation. Traditional methods of detecting moisture content and shrinkage ratio usually require a long measurement time, which is difficult to meet the needs of practical application. This investigated a rapid method based on a novel multispectral imaging system for acquiring multispectral images of samples in 25 wavebands over the spectral region between 675 and 975 nm at one time to detect moisture content and shrinkage ratio of dried carrot slices. The multispectral images of 600 carrot slice samples, which were dried at different times, were acquired using the multispectral imaging system. After extracting the spectral and GLCM features of the samples, prediction models were developed based on partial-least squares regression (PLSR) and least squares-support vector machines (LS-SVM) by using different feature combinations. Compared with PLSR models, LS-SVM models achieved better detection accuracy for moisture content and shrinkage ratio. The LS-SVM model obtained the following best results: coefficient of determination in prediction (Rp) = 0.942, root mean square error of prediction (RMSEP) = 0.0808%, and residual predictive deviation (RPD) = 2.636 for shrinkage ratio as well as Rp = 0.953, RMSEP = 0.0902%, and RPD = 3.271 for moisture content under static condition (without movement). The detection accuracy decreased with increasing movement speed of the test sample. When the movement speed of the sample was lower than 30 mm/s, the moisture content detected achieved satisfactory accuracy, with Rp = 0.941, RMSEP = 0.0981%, and RPD = 3.001. The novel multispectral imaging system shows potential for real-time detection of moisture and shrinkage of products during drying.

中文翻译：

---

多光谱成像系统快速检测胡萝卜干切片水分和收缩率

摘要 胡萝卜具有较高的营养价值和保健作用，深受消费者喜爱。对干燥胡萝卜片进行实时质量检测，使生产者可以及时调整干燥装置的工艺参数，从而保证最终产品质量，实现节能。传统的检测含水率和收缩率的方法通常需要较长的测量时间，难以满足实际应用的需要。研究了一种基于新型多光谱成像系统的快速方法，该方法用于在 675 到 975 nm 之间的光谱区域内一次获取 25 个波段的样品的多光谱图像，以检测干胡萝卜切片的水分含量和收缩率。600份不同时间干燥的胡萝卜切片样品的多光谱图像，使用多光谱成像系统获得。在提取样本的光谱和GLCM特征后，通过使用不同的特征组合，基于偏最小二乘回归（PLSR）和最小二乘支持向量机（LS-SVM）建立预测模型。与 PLSR 模型相比，LS-SVM 模型对水分含量和收缩率的检测精度更高。LS-SVM 模型获得了以下最佳结果：预测的决定系数 (Rp) = 0.942，预测的均方根误差 (RMSEP) = 0.0808%，收缩率的残余预测偏差 (RPD) = 2.636 Rp = 0.953、RMSEP = 0.0902% 和 RPD = 3.271 在静态条件下（无运动）的水分含量。检测精度随着试样移动速度的增加而降低。当样品移动速度低于30 mm/s时，水分含量检测达到满意的精度，Rp=0.941，RMSEP=0.0981%，RPD=3.001。新型多光谱成像系统显示了在干燥过程中实时检测产品水分和收缩的潜力。