干贝水分含量可视化检测方法的研究

摘要

水分含量影响干贝的口感、质地等品质特征,而且与其贮存期密切相关.应用高光谱成像与检测技术结合化学计量学方法,实现干贝水分含量的快速检测.实验采用高光谱成像系统采集380～1030 nm波段范围内的高光谱图像,采集得到6个不同干燥时期共90个干贝样本高光谱图像.提取所有样本感兴趣区域的平均光谱数据,采用连续投影算法(SPA)和权重回归系数法(B w)分别提取了7个和4个特征波长.基于所提取的特征波长和全波长分别建立光谱数据与水分含量的偏最小二乘回归(PLSR)模型,三种模型分别是SPA-PLSR,B w-PLSR和PLSR.建模集和预测集相关系数都高于0.95,预测均方根误差都低于10％,三种模型均获得了较好的预测效果,都能很好地预测干贝的水分含量.在所有模型中,SPA-PLSR模型具有较少的波长变量和较高的预测能力(97.28％),因此本文基于SPA-PLSR模型,采用伪彩色图像编程技术实现了干贝图像上每个像素点的水分含量的可视化预测.结果表明,高光谱成像技术结合特征波长提取算法可用于干贝水分含量分布的可视化检测.%Moisture content had a close relationship with the microbial growth and therefore it profoundly affected the dried scal-lop quality and safety,which is harmful to human health.In this study,a hyperspectral imaging system in the 380 ～1030 nm was used for rapid detection of the moisture content of scallops.90 hyperspectral images of six different dehydration periods were obtained.The mean spectra value of scallops from hyperspectral images were extracted and arranged in a matrix.Successive Pro-jection Algorithm (SPA)and Weighted Regression Coefficient (B w )were used to establish Partial Least Squares Regression (PLSR)models to correlate the spectral feature with moisture content.Seven and four optimal wavelengths were selected respec-tively to develop new simplified models called SPA-PLSR and B w-PLSR,which brought about sound prediction results (correla-tion coefficients were higher than 0.95,the root mean square error were less than 10％).Then the best model called SPA-PLSR with less wavelength variables and higher prediction of 97.28％,was applied to create visualization map to observe the distribu-tion of moisture content by pseudo color image programming technology.The results revealed the feasibility of hyperspectral im-aging technique combined with the optimal wavelengths for estimating and visualizing the moisture content of scallops.