随着人类生活质量的提高, 农产品重金属污染问题备受关注。 农作物中的重金属元素会通过食物链侵害人体健康, 而不同重金属元素对人体毒害差别较大, 因此农作物中含有重金属元素的类别识别至关重要。 传统重金属元素检测方法存在环节多、 耗时长、 成本高等缺点, 但高光谱遥感技术具有信息使用量大, 理化反演能力强, 分析速度快, 无损监测等优势, 逐渐成为农作物重金属污染分析的重要手段之一。 以不同CuSO₄·5H₂O和Pb(NO₃)₂浓度梯度土壤胁迫下典型农作物玉米生长的叶片光谱为研究对象, 引入光谱包络线去除(CR)、 光谱比值(SR)、 分数阶微分(FOD)同时结合改进红边比值指数(MSR)构建铜铅元素识别指数(CLI); 通过挑选与铜铅元素种类相关性最强的三个分数阶微分阶数的CLI值建立铜铅元素判别特征点(CLDFP); 再利用欧式聚类(EC)将训练集样本分为铜污染与铅污染两类并结合圆心连线的垂直平分线(PB), 建立基于EC-PB识别铜铅元素种类的二维坐标系下判别规则线(CLDRL)和三维坐标系下判别规则面(CLDRP), 从而实现玉米叶片光谱信息的重金属铜铅元素种类准确识别。 研究结果表明, CR-SR-FOD光谱变换处理增加了玉米叶片光谱信息与铜铅元素种类之间的相关性; 各阶次FOD对应的CLI与铜铅元素种类相关系数各不相同, 随着阶次的增加, 相关性呈现先递增后递减的趋势, 其中相关系数最高的三个阶次分别为1.2阶, 0.7阶, 1.0阶; 在二维坐标系下训练集样本判别正确率为78.95%, 验证集样本判别正确率为75.0%; 在三维坐标系下训练集样本判别正确率为76.32%, 验证集样本判别正确率为75.0%, 证明了基于EC-PB构建的二维CLDRL和三维CLDRP光谱判别规则可以有效识别玉米叶片中铜铅污染元素种类。

Heavy metal pollution of agricultural products has attracted much attention along with the improved human quality of life. The heavy metal elements in crops will harm human health through the food chain, and different heavy metal elements have a large difference in toxicity to the human body. Therefore, it is crucial to distinguish the types of heavy metal elements in crops. There are many shortcomings in the traditional methods of detecting heavy metals such as many links, long time, and high cost. However, hyperspectral remote sensing technology has the advantages of abundant information usage, strong physical and chemical inversion capabilities, fast analysis speed, non-destructive monitoring and so on. It has gradually become one of the important methods for analysing heavy metal pollution in crops.Taking the leaf spectra of a typical corn crop growing under soil stressed by different CuSO₄·5H₂O and Pb(NO₃)₂ concentration gradients as the research object, the copper (Cu) and lead (Pb) identification index (CLI) was builtbased on spectral processing results of continuum removal (CR), spectral ratio (SR)and fractional-order derivative (FOD) combining with modified red edge simple ratio index (MSR). Then the Cu and Pb element discrimination feature points (CLDFP) were established by selecting the three CLI values of fractional differential orders that have the strongest correlation with the types of Cu and Pb elements. And then, the Cu and Pb elements discriminant rule line (CLDRL) under the two-dimensional coordinate system (2D) and the discriminant rule plane (CLDRP) under the three-dimensional coordinate system (3D) were structured to identify the types of Cu and Pb elements. Based on the Euclidean cluster (EC)- the perpendicular bisector (PB) by using the EC to divide the training samples into two sets of Cu pollution and Pb pollution and combining with the PB to connect the circle enters the sets so that the types could be accurately identified on the heavy metal Cu and Pb elements in the spectral information of corn leaves. The results showed that the correlation between the spectral information of corn leaves and the types of Cu and Pb elements was enhanced because of the CR-SR-FOD spectral transformation processing. The correlation coefficients of the CLI corresponding to each order of FOD and the types of Cu and Pb elements were different. With the increase of orders, the correlation showed a trend of increasing first and then decreasing. Among them, the three values of orders of the highest correlation coefficients were 1.2, 0.7, and 1.0 respectively. The accuracy rate of the training set samples was 78.95% andthe accuracy rate of the verification set samples was 75.0% when discriminated under the 2D, and the accuracy rate of the training set samples was 76.32% and the accuracy rate of the verification set samples was 75.0% when discriminated under the 3D, it is proved that the spectral discriminant rulesof 2D CLDRL and 3D CLDRP based on EC-PB could effectively identify the types of Cu and Pb pollution elements when they polluted the corn leaves.