基于多频信息融合的探地雷达林木根系原位探测识别研究

Against to the technical demand for the forest underground biomass estimation and roots measurement in situ, A new methods of tree roots detection and recognition in situ using ground-penetrating radar(GPR) under forest soil conditions that based on multi-frequency information fusion is presented in this project. Combination with the complex structure characteristics of forest soil,such as slash,rocks,buried organic matter,rodent tunnels,etc. The principle of the new roots recognition method using ground penetrating radar, and the key science questions are thorough researched. Including,GPR image quality objective assessment modeling based on gradient and entropy function, the method for optimal antenna frequency selection, clutter suppression technique based on wavelet transform, target recognition technique based on multi-frequency image fusion, subpixel location method of roots based on image moments,the root diameter estimation technique based on artifical neural network,etc. So as to achieve the precision recognition of root morphology under forest soil conditions. The project concerns multidisciplinary knowledge fusion, sunch as GPR image acquisition and optimal antenna frequency selection methods, image quality assessment theory, pattern recognition technique, fuzzy set theory, image processing technique and graph visualization,etc. The research has great significance to the nondestructive detection and morphology analysis of roots, the forest underground biomass estimation and the effect research of the root to the forest carbon sequestration.

针对森林地下生物量估算及林木根系原位观测和测量技术需求，本项目拟提出一种森林土壤介质下基于多频信息融合的探地雷达林木根系原位探测识别新方法。结合森林土壤介质复杂结构（含石砾、动物孔穴、腐烂根、腐殖质等）特征，对探地雷达根系识别新方法的机理及关键科学问题进行深入研究，包括基于梯度函数和熵函数的探地雷达图像质量无参考评价模型的建立及天线频率优选方法、基于小波变换的杂波抑制技术、基于多频图像信息融合的目标识别技术、基于图像矩的根系位置亚像素定位方法、基于人工神经网络的根径估计方法等，以期实现森林土壤介质下林木根系形态的精确识别。项目研究内容涉及探地雷达根系图像获取技术及天线频率优选方法、图像质量评价理论、模式识别技术、模糊集合理论、图像处理技术、图形可视化等多学科领域知识的融合。该方法的研究对根系无损探测及形态分析、森林地下生物量计算以及根系对林业碳汇作用的研究具有重要意义。

根系是树木生长于地下的营养器官，它具有吸收、固着、输导、合成、储藏和繁殖等功能；对于林木根系形态特征研究对树木、土壤、乃至对整个生态系统的水力平衡以及营养循环均具有重要意义。项目组针对林木根系原位无损观测和测量技术需求，依据项目研制计划以及项目具体实施情况，积极开展相关仿真、试验和算法研究工作，全面完成了项目各项研究内容。本研究以探地雷达为探测手段，通过基于空间、频谱和梯度熵值图像质量无参考评价技术研究实现了林木根系GPR图像质量客观评价与天线频率优选；采用基于林木根系GPR图像小波变换技术完成了林木根系B-SCAN图像杂波抑制；通过多尺度图像融合技术进行多频图像融合提高根系回波信号强度与图像边缘检测精度；提出了基于拉格朗日乘数法离散函数求导和导数加权的思路优化基于Hough变换的波速估计为项目研究提供精准电磁波波速参数；深入研究并优化基于波动方程理论和Stolt插值技术林木GPR图像二维与三维合成孔径反演成像算，实现林木根系原位无损形态测量。本研究旨在提供一种探地雷达林木根系形态原位探测识别新方法，拟为林木根系形态分析、森林地下生物量计算、森林碳汇计量以及林木根系构型定量分析提供技术支撑，具有较为重要研究参考价值与广阔的应用前景。