Skip to content
YIZE
P.004 / PROJECT2025 — 2026

根系动态生长表型｜测量方法与发明专利Root Growth Phenotyping · Measurement System and Invention Patent

Root Growth Phenotyping · Measurement System and Invention Patent根系动态生长表型｜测量方法与发明专利

一项已授权（不是受理中）的发明专利：从时序图像高通量测量根系动态生长——时序差分、骨架提取、最短路径定长，最终长成生长曲线。

An invention patent — granted, not pending — for a high-throughput method that measures root length over time from image sequences: temporal differencing, skeleton extraction and shortest-path rooting, assembled into growth curves.

角色Role
发明人之一 · 测量软件独立完成Inventor · measurement software built independently
类型Type
科研Research
年份Year
2025 — 2026
  • Computer vision
  • Image calibration
  • Skeleton extraction
  • Python
Research Figure
P.004-A
Research Figure · placeholder
目录Contents
  1. 一个在可见变化之前就发生的响应
  2. 方法，四步
  3. 专利：已授权
  4. 它为什么存在
  5. 这一页不写的

一个在可见变化之前就发生的响应

地上部还看不出任何异样的时候，根系已经开始对水分亏缺作出反应了。所以根系生长动态是抗旱研究里少数几个直接有用的表型之一。它同时也是最难伺候的：根看不见、缠在一起，而在高通量试验里是以千为单位被拍下来的——人工测量既扩不了规模，也重复不出来。

这个项目做的就是把它量出来。专利方法在一套高通量根系表型装置的时序图像上工作，把每个植株的主根长度随时间的变化算成一条生长曲线，全程不碰植株。

方法，四步

  1. 标定。 从参考图像生成一套标定指令集，用它对序列中的每一帧做校正——这样第 1 帧里的一个像素和第 400 帧里的一个像素指的是同一件事。
  2. 差分并累加。 校正后的子图像序列做时序差分，差分结果按像素累加，得到观测周期内的一张静态累加图像。
  3. 分割并细化。 自适应二值化把根系分离出来，再提取单像素宽度的中心线骨架。
  4. 测量。 在骨架上做最短路径寻路，得到主根长度。在每个时间点重复这一步，就长出了生长曲线——那个静态图像给不了的动态量。

这套设计真正在意的是：每一步都是在写明的输入上的一次可记录、可重跑的操作。复现不出来的测量是轶事，这个是可以重跑的一条管线。

专利：已授权

一种高通量植物根系动态生长表型的测量方法及系统——授权公告日 2026 年 8 月 7 日（专利号 ZL 2025 1 1700810.2，证书号 第9167770号，授权公告号 CN 121504884 B），专利权人为中国农业大学。九位发明人，我列第三。实现这套方法的测量软件由我独立完成，从第一行到最后一行。

它为什么存在

这项工作服务于一个国家级课题——玉米根系表型响应水分亏缺的监测模拟与高水效靶向调控机制研究。该课题要沿干旱梯度拿到根系动态，规模是人工测量达不到的：它创新性地搭了根盒—盆栽—大田相结合的多尺度研究体系，用高光谱定量解析不同干旱梯度下根系光谱响应的动态特征，整合多组学数据揭示代谢遗传机制，再基于多模态数据融合与机器学习建立模拟模型。本系统产出的生长曲线，是这些「表型—代谢—调控」分析的输入，而不是一组孤立的数字。

这一页不写的

专利的权利要求全文与附图（它们已经公开，按专利号就能查到，抄在这里不增加任何信息），以及课题的试验数据——那属于课题。

目录Contents
  1. Overview
  2. The method
  3. The patent
  4. Why it exists
  5. What is not on this page

Overview

Root systems respond to water deficit before anything visible happens above ground, which makes root growth dynamics one of the directly useful phenotypes in drought research. It is also one of the awkward ones: roots are invisible, entangled, and — in high-throughput trials — photographed by the thousand, so manual measurement neither scales nor repeats.

This project measures it. The patented method takes a time series of root images from a high-throughput phenotyping setup and turns it into a growth curve of main-root length over time, per plant, without touching the plant.

The method

  1. Calibrate. A calibration instruction set is generated from reference images, and every frame in the sequence is corrected against it, so that a pixel means the same thing in frame 1 and frame 400.
  2. Difference and accumulate. The corrected sub-image sequences are processed with temporal differencing; the differenced frames are accumulated pixel-wise into a single static image of the observation period.
  3. Segment and skeletonise. Adaptive binarisation isolates the root; a single-pixel-wide centreline skeleton is extracted from it.
  4. Measure. A shortest-path algorithm over the skeleton yields the main root length. Repeating this at each time point produces the growth curve — the dynamic quantity the static image cannot give.

The point of the design is that every step is a recorded, re-runnable operation on a declared input. A measurement that cannot be re-derived is an anecdote; this one is a pipeline.

The patent

一种高通量植物根系动态生长表型的测量方法及系统 — granted 7 August 2026 (专利号 ZL 2025 1 1700810.2, 证书号 第9167770号), applicant and patent holder 中国农业大学; nine inventors, of which I am the third. The measurement software implementing the method was built by me independently, end to end.

Why it exists

The work serves a national programme on maize root responses to water deficit — 根系表型响应水分亏缺的监测模拟与高水效靶向调控机制研究 — which needs root dynamics across drought gradients at a scale manual measurement cannot reach. The growth curves this system produces are the input to that programme's phenotype–metabolism–regulation analyses, not a standalone gallery of numbers.

What is not on this page

The patent's full claims and figures (they are published and findable by the patent number, and repeating them here adds nothing), and the programme's trial data, which belongs to the programme.