基于微根管图像的根系形态特征快速提取技术

Rapid extraction technology of the root morphological characteristics via minirhizotron image

林木根系的形态和分布特征能够反映林木生长状态和土壤水养资源响应及利用机制,对于提升人工林培育质量,发挥人工林水土保持作用具有重要意义。利用微根管技术并对毛白杨根系图像进行剪裁、标注和滤波处理,运用大津法、K均值聚类算法、模糊C均值聚类法(FCM法)3种图像处理方法分割根系结构,并以14 d为试验采集周期,获得林木根系分布特征。结果表明,FCM法的正确率达到90%,对于根系的分割性能最优。通过根系形态特征参数提取,以“根总长”为例,分析不同滴灌处理下根系的生长周转情况,表现出良好的规律性,该技术的开发为根系生态学的研究提供了一种无损且高效的研究方法。

[Background]The morphology of roots and the distribution of soil layers reflect the growth state and utilization strategy of soil water-trophic resources,it has a great significance to improve the quality of plantation cultivation and soil and water conservation.However,the traditional root research methods,such as soil drill method,soil column method,have complex test processes and cause certain damages to the root system.Therefore,it is very important to propose a new experiment and research method that may achieve nondestructive,automatic and digital monitoring of root growth state.[Methods]Selecting 3 healthy trees(Populus tomentosa),designing the experiment based on the minirhizotron technology,and taking 14 days as the data collection period,the root distribution in different soil layers was obtained.Root image was clipped and the true value was obtained via manual annotation.After image filtering processing,the Otsu,K-means,and FCM(fuzzy C-means clustering algorithm)methods were used to segment images.The correct rate was selected as the quantitative analysis index,and the experiment results were obtained.[Results]Through the qualitative analysis,there was over-segmentation when Otsu method was used to have image segmentation for the two types of root,and over-segmentation and under segmentation for the K-means method.The segmentation results via FCM method for the two types of roots were similar to the one via manual annotation.Through quantitative analysis,correct rate by K-means and FCM method were similar,all better than that by Otsu method.The accuracy of FCM method reached 90%.[Conclusions]The experiment was designed and was in steady operation,and the changes of root systems were obtained under different soil layers,i.e.,the nondestructive monitoring was achieved.Through image segmentation,the segmentation algorithm was obtained by which its average error rate was<10%,i.e.,the automatic monitoring was achieved.The characteristic parameters such as total root length was taken as example,the root growth turnover under different drip irrigation strategies was analyzed,and showed good applicability,which offered an efficient and nondestructive approach for the root ecology research.