Liu Yao, Qiao Fu, Wang Shuwen, Wang Runtao, Xu Lele
School of Electronic and Electrical Engineering, Lingnan Normal University 29 Cunjin Road, Chikan District Zhanjiang 524048 Guangdong Province China
School of Computer Science and Intelligence Education, Lingnan Normal University Zhanjiang 524048 China.
RSC Adv. 2021 Nov 15;11(54):33939-33951. doi: 10.1039/d1ra03664e. eCollection 2021 Oct 18.
Human beings are confronted with a serious health hazard when ingesting contaminated with heavy metals, and thus it is significantly necessary to identify heavy metal contaminated . This study investigates the feasibility of hyperspectral imaging to identify heavy metal contamination in rapidly. To reduce the effects of noise, four different spectral pretreatments were performed on the original spectra. To select characteristic wavebands for identification, four waveband selection algorithms based on neighbourhood rough set theory were proposed, namely, mutual information, consistency measure, dependency measure, and variable precision. The selected wavebands were input to an extreme learning machine to construct classification models. The results demonstrated that multiplicative scatter correction pretreatment was suitable for hyperspectral imaging datasets. The identification models exhibited satisfactory performance to distinguish healthy from those contaminated by both individual and multiple heavy metals. The identification results of Cd and Pb contaminated samples were more accurate than those of Cu and Zn contaminated samples. When the number of training samples decreased the identification performance decreased, but not significantly. The results showed that combined with pattern recognition analysis hyperspectral imaging technology can be used to distinguish healthy samples from those contaminated by heavy metals, even with only a small number of training samples. This model is suitable for applications in analysing many shellfish rapidly and non-destructively.
人类在摄入受重金属污染的[具体物质未明确，推测为贝类等]时面临严重的健康危害，因此识别受重金属污染的[具体物质未明确，推测为贝类等]非常必要。本研究探讨了高光谱成像快速识别[具体物质未明确，推测为贝类等]中重金属污染的可行性。为降低噪声影响，对原始光谱进行了四种不同的光谱预处理。为选择用于识别的特征波段，提出了基于邻域粗糙集理论的四种波段选择算法，即互信息、一致性度量、依赖性度量和变精度。将所选波段输入极限学习机以构建分类模型。结果表明，乘法散射校正预处理适用于[具体物质未明确，推测为贝类等]高光谱成像数据集。识别模型在区分健康的[具体物质未明确，推测为贝类等]与受单一和多种重金属污染的[具体物质未明确，推测为贝类等]方面表现出令人满意的性能。镉和铅污染样本的识别结果比铜和锌污染样本的更准确。当训练样本数量减少时，识别性能下降，但不显著。结果表明，结合模式识别分析，高光谱成像技术可用于区分健康的[具体物质未明确，推测为贝类等]样本与受重金属污染的样本，即使只有少量训练样本。该模型适用于快速、无损地分析大量[具体物质未明确，推测为贝类等]的应用。
