一种基于多模态图像技术的葡萄水分胁迫识别

提出了一种利用多模态图像技术,以实现被葡萄水分胁迫水平的测定方法,通过检测获取葡萄植株表面图像的反射率和纹理信息与水分胁迫水平之间的关系,从而实现植物缺水报警。试验将盆栽葡萄人为建立不同的水分胁迫水平,利用3CCD照相机(三通道的R,G和IR)、多光谱相机(在900,970 nm的光谱波段覆盖)和一个数字彩色摄像机(RGB)对叶片定期进行监测。试验采用偏最小二乘(PLS)方法预测水含量的纹理特征和光谱特性,在葡萄生长的前期,RGB相机获得的纹理参量对含水量预测结果的rp,RMSEP和偏差值分别为0.77,1.15和-0.14,而利用3CCD相机获取的反射参量对含水量预测结果的rp,RMSEP和偏差值分别为0.77,1.22和-0.26;在葡萄生长后期,RGB相机获取的纹理参量对含水量预测结果的rp,RMSEP和偏差值分别为0.81,1.34和0.26,而利用3CCD相机获取的反射参量对含水量预测结果的rp,RMSEP和偏差值分别为0.74,1.46和0.15。通过监测植株覆盖率与不同水分灌溉植株的生长周期发现,植株覆盖率能对葡萄植株的水分胁迫检测做辅助参考变量。试验结果表明,所设计的多传感器系统可用于支持葡萄水分胁迫检测的决策,有利于葡萄的田间管理。

Water Stress Identification in Grape Vines Based on Multi-modal Image System

A multi-modal sensor system was designed to measure the reflectance and textural signature of grape plant surfaces and to identify different water stress levels so as to alarm and avoid the water deficiency.Several levels of water treatment were created for the potted grape vine in the field.The leaves were periodically monitored with a 3CCD camera （three channels in R,G,and IR）,a multispectral camera （covering the spectral bands at 900,970 nm）,and a digital color camera （RGB）.The partial least squares （PLS）method was used to predict water content by using textural features and reflectance features.For the textural features,the model was optimized with three latent variables （LVs）,it has coefficient of prediction （rp）,root mean square error of prediction （RMSEP）,and bias of 0.77,1.15,and-0.14 respectively for grape vine in early stage and 0.81,1.34,and 0.26 in late stage.For the reflectance features,the rp,RMSEP,and bias for the optimal PLS models were 0.77,1.22,and-0.26 respectively for grape vine in early stage,and 0.74,1.46,and 0.15 in late stage.The experimental results indicated that the designed multi-modal sensor system could be used to support decision-making for grape vine water stress detection and be beneficial for grape field management.