重金属的快速精准检测对农作物生长监测、品质控制具有重要意义。项目以重金属（Cd、Pb、Cd与Pb复合）胁迫下的油菜为载体，研究不同重金属胁迫下，重金属离子对荧光高光谱图像的影响及重金属络合物含量变化对可见光-近红外高光谱图像的作用关系，探讨重金属类别检测机理；在重金属不同梯度水平胁迫下，分析油菜叶片内部生理生化指标、组织结构、外部表面微观结构变化及荧光-可见光-近红外多源高光谱图像的响应规律，探究重金属含量无损检测机理。确定重金属的有效敏感波段和图像特征，进行重金属类别鉴别。提取重金属不同水平胁迫下多源高光谱图像的深度特征，消除复合重金属间交互作用干扰的影响，优化复合重金属多源光谱、图像深度特征组合。利用信息学技术及回归算法构建油菜叶片重金属高精度检测模型，进行验证试验。本项目旨在提出一种基于多源高光谱图像深度特征的油菜叶片重金属高精度无损检测方法，突破了传统方法普适性差、精度低的局限性。

The rapid and accurate detection of heavy metals is of great significance for crop growth monitoring and quality control. In this project, oil crop rape will be chosen as carrier of different heavy metals including Cd, Pb, Cd and Pb composite. In different kinds of heavy metals, the relationship between heavy metal ions and fluorescent hyperspectral images, and the relationship between heavy metal complexes and visible-near infrared hyperspectral images will be studied, and the detection mechanism of heavy metals’ category. Under the stress of different gradient levels of heavy metals, the changes of fluorescence-visible-near infrared multi-source hyperspectral images caused by the changes of biochemical indexes, tissue structure, external surface microstructure will be studied, and exploring the non-destructive testing mechanism of heavy metal contents. The effective sensitive bands and spectral image features of each heavy metal were determined to identify heavy metal species. The depth characteristics of fluorescence/visible-near infrared multi-source hyperspectral images under different levels of heavy metals stress were extracted, and the interaction interference between composite heavy metals was excluded, and the combination of multi-source spectra and image depth features of heavy metals was optimized. Using informatics technology and regression algorithm to construct a high-precision detection model of heavy metals in rape leaves, and carry out verification experiments. This project aims to propose a high-precision non-destructive detection method for heavy metals in rape leaves based on the depth characteristics of multi-source hyperspectral images, which breaks through the limitations of traditional methods for detecting poor universality and low precision.