面向高通量植物表型测量的多模态图像结构化协同分析

植物图像分析是制约高通量植物表型测量发展的技术瓶颈。现有方法普遍缺乏对复杂植物结构进行建模分析的有效手段、且只能处理单一模态的植物图像，无法满足实际应用对表型参数精确性和多样性的需求。本项目拟面向高通量植物表型测量的实际应用，针对最典型的两类模式植，即阔叶植物和线状植物，围绕着单模态图像结构化分析和多模态图像协同分析两个方面的基础性科学问题开展研究。首先，建立多模态植物图像结构化协同分析的总体框架；其次，针对阔叶植物叶片实例分割、线状植物骨架结构重建、多模态植物结构对应、多模态植物图像视觉协同显著性分析等四个具体问题进行研究，提出有效的算法解决方案；最后，通过实际数据对算法进行性能评估，进一步优化完善所提出的算法。本项目将发展植物图像结构化协同分析的理论和方法，具有一定的学术研究价值；同时，将为高通量植物表型测量提供基础技术支撑，具有重要的应用价值和广阔的应用前景。

Plant image analysis is widely considered to be the bottleneck towards high-throughput plant phenotyping. Existing techniques commonly lack of effective measures to model and analyze the very complex structures of plants, and moreover, they mostly can only deal with one single modality of plant images, which thereby fail to satisfy the demands on the precision and diversity of plant phenotyping in realistic applications. Towards this end, this project aims to focus on two most typical model plants, i.e., broad-leaved plants and curvilinear plants, and study two aspects of fundamental problems in plant image analysis, i.e., structural analysis and multi-modal co-analysis. Firstly, an overall framework for co-analysis of multi-modal plant images will be established. Secondly, four particular problems will be studied in detail and new algorithms will be proposed accordingly, including leaf instance segmentation of broad-leaved plants, skeleton structure reconstruction of curvilinear plants, structure correspondence across multiple modalities and co-saliency analysis of multi-modal plant images. Finally, experiments on real dataset will be carried out to evaluate and further refine the proposed algorithms. This project will academically contribute by developing the theory and methodology of structural co-analysis of multi-modal plant images. Furthermore, it will also contribute practically by providing effective algorithms to enable the very wide applications of high-throughput plant phenotyping.