基于阈值选取高光谱感兴趣区域的牛肉掺假检测

Detection of adulterated beef based on selecting hyperspectral region of interest with threshold

中文关键词:  高光谱成像技术  阈值  感兴趣区域  牛肉  掺假  猪肉  偏最小二乘法

英文关键词:hyperspectral imaging technology  threshold  regions of interest  beef  adulteration  pork  partial least squares regression

中文摘要:

      目的 建立一种通过阈值选取高光谱图感兴趣区域建立模型从而提高快速检测牛肉掺假猪肉水平的方法。方法 以2%为间隔, 掺假比例为2%~50% (w/w), 配制掺入猪肉糜的掺假牛肉样本并采集高光谱反射数据, 根据某波长处光谱数据频率曲线设定阈值提取高光谱感兴趣区域, 同时以直径为150个像素的圆形区域提取光谱作为对比。采用浓度梯度法、Kennard-Stone法、光谱-理化值共生距离法(sample set partitioning based on joint x-y distance, SPXY)、极大线性无关法将样本划分为校正集和预测集, 借助多元散射校正结合平滑预处理后采用偏最小二乘法建立掺假预测模型进行比较分析。结果 通过设定阈值提取感兴趣区域建立的模型比圆形区域提取所建模型更加稳定、精准, 其中SPXY法划分样本所建立的模型预测效果最好, 校正相关系数 为0.9733, 验证集相关系数 为0.9515。结论 基于高光谱技术通过设定阈值提取特征光谱并结合化学计量学可提高预测牛肉掺假的能力。

英文摘要:

      Objective To establish a method for improving the rapid detection of beef adulterated with pork by setting threshold to select hyperspectral imaging of region of interest (ROIs). Methods Mince beef samples were adulterated with minced pork in the range 2%–50% (w/w) at approximately 2% intervals. Hyperspectral images were acquired in the reflectance mode. Then, threshold was set by frequency curve of spectral data to extract ROIs. Meantime, circular region with 150 pixels in diameter was used for comparision. Then, spectral data were divided into the calibration set and validation set by concentration gradient method, Kennard-Stoned, sample set partitioning based on joint x-y distance (SPXY) and maximal linear independent method. The partial least squares regression (PLSR) model was established and preprocessed by multiple scattering correction and smoothing for comparison and analysis. Results The prediction model was more stable and accurate by using threshold to extract ROI than circular region. The model preprocessed by SPXY was the best with the coefficient correlation of calibration set ( ) of 0.9733 and the correlation coefficient of validation set ( ) of 0.9515. Conclusion The method of extracting characteristic spectra by setting threshold and combining with chemometrics in near-infrared (NIR) hyperspectral imaging can improve prediction ability of adulterated beef.