[目的]通过近红外光谱技术解决贻贝重金属铅污染问题。[方法]应用近红外反射光谱结合模式识别的方法进行重金属铅污染检测。首先获得了在950～1 700 nm范围内的健康贻贝和重金属铅污染贻贝光谱数据，应用基于随机变量组合的变量重要性分析(variable importance analysis based on random variable combination，VIAVC)波段选择算法对光谱数据降维，筛选最佳波段子集。针对检测健康贻贝和重金属铅污染贻贝是一个不平衡的分类问题，研究探索一种基于万有引力的固定半径最近邻(gravitational fixed radius nearest neighbor，GFRNN)方法用于贝类重金属铅污染识别。[结果]相较于传统的K最近邻法、固定半径近邻法和支持向量机算法，研究提出的VIAVC-GFRNN方法在检测重金属铅污染方面表现出更优异的性能，并且不受样本不平衡率的影响。VIAVC-GFRNN模型的接收者操作特征曲线下面积值达到了0。988 6，检测精度和几何均值均达99。17%。[结论]近红外光谱结合模式识别方法在检测贻贝中铅污染方面具有很大的潜力。
Identification of heavy metal Pb pollution in Perna viridis based on near-infrared spectroscopy
[Objective]Addressing the heavy metal lead pollution in oysters using near-infrared spectroscopy technology.[Methods]This study proposed the use of near-infrared reflectance spectroscopy combined with pattern recognition for detecting Pb contamination.Initially,spectral data of healthy mussels and Pb-contaminated mussels in the range of 950～1 700 nm were collected.The wavelength selection algorithm of variable importance analysis based on the random variable combination(VIAVC)was utilized to reduce the dimensionality,and selected the optimal subset of wavelengths.Considering the detection of healthy mussels and Pb-contaminated mussels as an imbalanced classification problem,the gravitational fixed radius nearest neighbor(GFRNN)method based on universal gravity was explored for identifying Pb contamination in mussels.[Results]The experimental results demonstrated that the proposed VIAVC-GFRNN method outperformed traditional algorithms such as K-nearest neighbor,fixed radius nearest neighbor,and support vector machine algorithms in detecting Pb contamination,while remaining unaffected by the imbalance ratio.The area under the receiver operation curve value of the VIAVC-GFRNN model reached 0.988 6,with a detection accuracy and geometric mean of 99.17%.[Conclusion]Near-infrared spectroscopy combined with pattern recognition methods has great potential for detecting Pd pollution in mussels.
near-infrared spectroscopymusselsheavy metal detectionunbalanced classification
国家科技期刊平台
NSTL
万方数据
