针对土壤遮挡时根系图像信息不全的问题, 提出一种热红外成像根系表型检测方法, 结合Criminisi改进算法实现根系图像信息的增强和修复, 并研究玉米根系表型与种子活力之间的关系。 首先, 设计一种适应于玉米根系构型的环形双层石英培养装置迫使玉米根系贴壁生长, 分别将老化0, 1, 3和6 d的玉米种子种植在环形培养装置中。 基于水和土壤比热容具有显著差异的特点, 利用水对玉米苗根茎进行滴灌, 并通过热空气对培养装置中的玉米根系进行短时热激励, 再用红外热像仪采集根系红外热像, 利用土壤与根土间隙水流温度的差异实现土壤遮挡处根系的热成像。 其次, 对预处理后的根系热像, 进行端点和最佳匹配对判定, 并利用Criminisi改进算法对红外热像中的断根连接, 实现根系热红外图像的修补。 最后, 利用以上方法分别对不同老化天数的玉米种子幼苗进行根系表型检测验证。 结果表明, 所提出的热红外成像方法可有助于土壤遮挡处根系的表型图像信息增强, 比彩色图像提取的根系表型参数精度提高约0.5%~10%。 玉米种子老化1d后其根系表型参数总根长（RTL）和总根数（RTN）未见明显差异, 但老化3d和6d的种子其根系表型参数具有显著差异, RTL减少20%～35%, RTN减少10%～55%, 反映了玉米种子长时间老化后其活力存在显著下降。 不同老化天数的玉米根系表型参数RTL和RTN均与老化天数呈显著负相关, 可作为种子活力的重要指标参数, 其中, 种子根系RTN参数对老化更为敏感, 更能够直观反映种子的活力水平, 老化1和3 d的种子发根与未老化种子相比均推迟1 d; 老化6 d的种子其发根则推迟2 d, 且后续根系发育一直迟缓。 所提的基于热红外成像的根系表型检测结合Criminisi改进算法的根系表型检测方法, 可用于作物根系表型高通量无损检测, 具有广阔的应用前景。

Aiming at the problem of incomplete root image information because of blocking by the soil, the paper proposed a root phenotypic method by using thermal image combined with improved Criminisi algorithm for root image repair and studied the relationship between the root phenotype and seed vigor. First, an annular double-layer quartz culture device adapted to maize root configuration was designed to push maize roots to grow along the inner wall of the device, and the maize seeds aged 0, 1, 3 and 6 d were planted in the annular culture device respectively. Base on the significant difference of heat capacity between soil and water, water was used to irrigate the seedling along their stems followed by short-time hot air thermal excitation, and then infrared thermal images were captured based on the temperature difference between the soil and interstitial water flow around the roots. Secondly, the endpoints of the root thermal images after preprocessed were selected and matched for connecting using improved Criminisi algorithm to repair the root image. Finally, different aged-day maize seeds were applied for seeding root phenotyping detection to verify the mentioned method which results show that the proposed thermal infrared imaging method can help to enhance the root phenotypic image information which improves the precision of phenotypic parameters about 0.5%~10% comparedwith color image. The was no significant difference of Root Total Length (RTL) and Root Total Number (RTN) after 1 d aging, but there was remarkable difference of RTL and RTN after 3 and 6 d aging which decreased about 20%~35% and 10%~55% respectively. In general, the maize root phenotypic parameters such as RTN and RTL were significantly negative with the aging-day, which can be used as important index parameters of seed vigor. Furthermore, RTN is more sensitive to impress a seed vigor. Root number of 1 d/3 d and 6 d aging days increasing delayed about 1day and 2 day compared with 0 aging-day seeds respectively. The proposed root phenotypic detection method based on the thermal infrared imaging combined with improved Criminisi algorithm for root image repair can be used in root high throughput non-destructive detection, which has a broad application prospect.