基于低空遥感与GA-BP神经网络的葡萄叶片含水量估算研究

Research on Estimation of Water Content in Grape Leaves Based on Low-altitude Remote Sensing and GA-BP Neural Network

以武威市凉州区威龙庄园的葡萄为研究对象,使用无人机搭载多光谱相机对田间葡萄叶片含水量进行无损检测。选用Pix4D mapper软件对多光谱影像进行拼接,并利用矫正板矫正,通过指数计算器得到5个波段的光谱反射率影像。随机选取70组葡萄叶片光谱反射率作为训练样本,采用经典BP神经网络建立基于多光谱图像的葡萄叶片含水率模型,并融合遗传算法(Genetic Algorithms,GA)优化神经网络,输入量为蓝、绿、红、红边、近红等5个波段图像对应反射率,输出为葡萄叶片含水量。30组葡萄叶片验证样本用于模型反演数据的相关性分析。试验结果表明,利用多光谱图像信息并结合遗传算法优化的BP神经网络葡萄叶片含水量的反演模型,多光谱反演的含水量模型拟合相关系数达0.76982,30组验证集中葡萄叶片含水率实测参考值和网络反演值的相关系数r为0.8146,反演结果比较理想。本方法可实现对葡萄叶片含水量的快速无损准确检测,有助于西北干旱区农业的灌溉决策与精准管理。

Taking the grapes of Weilong Manor in Liangzhou District of Wuwei City as the research object,the water content of grape leaves in the field went through non-destructive testing using UAV-mounted multispectral cameras.Pix4D mapper software was used to stitch the multispectral images,and a correction plate was used to correct them,with spectral reflectance images of 5 bandsobtained through an index calculator.70 groups of grape leaf spectral reflectance were randomly selected as training samples,and the classical BP neural network was used to establish the water content model of grape leaves based on multi-spectral images,with Genetic Algorithms(GA)used to optimize the neural network,the inputsbeing five-band blue and green,red,red edgeand near-redimages corresponding to reflectanceand the output being the water content of grape leaves.30 groups of grape leaf validation samples were used for correlation analysis of model inversion data.The experimental results show that using the multi-spectral image information combined with the genetic algorithm to optimize the BP neural network water content inversion model of grape leaves,the water content model fitting correlation coefficient of multi-spectral inversion reached 0.76982,andthe correlation coefficient r of the measured reference value of grape leaf water content and the network inversion value in the 30 sets of validation setswas 0.8146,indicating an ideal inversion result.This method can realize the rapid non-destructive and accurate detection of the water content of grape leaves,which is helpful for the irrigation decision-making and precise management of agriculture in the northwest arid region.