高光谱成像的水稻螟虫蛀入检测方法

Research on Rich Borer Detection Methods Based on Hyperspectral Imaging Technology

为了控制水稻螟虫预警和喷洒农药用量,实现对水稻螟虫虫害的无损检测,提出了基于主成分分析特征波段检测方法和基于迭代阈值的最优波段检测方法,确定了水稻茎秆螟虫检测的特征波段和最优波段,提取出单波段和组合波段的图像来分割虫孔,从而实现水稻螟虫的精准的无损检测。首先通过高光谱得到的120个样品反射率信息分析确定了光谱区域为450~1000 nm。基于主成分分析特征波段检测方法,对高光谱图像进行主成分分析,通过前五个主成分图像比较确定第三主成分图像为最佳,然后根据第三主成分图像中各个波段的贡献率来选取特征波长(668.8和750 nm),最后结合全局阈值分割和图像掩膜等图像处理方法实现对虫孔区域的判别。而利用基于迭代阈值的最优波段检测方法,在可见光波段450~750 nm范围和近红外波段750~1000 nm范围内应用混合距离挑选最佳的单波段,通过单波段来确定组合波段,对单波段和组合波段进行迭代阈值分割,其中753.5 nm波长分割效果最好,故确定753.5 nm为最优波长,然后提取该波长的图像采用一种基于迭代阈值虫孔提取方法和形态学处理,最后能对水稻茎秆虫孔区域进行判别来实现水稻茎秆虫害是否存在。对60个虫害水稻茎秆和60个正常水稻茎秆进行检测,应用基于主成分分析特征波段检测方法在668.8和750 nm波长处检测率分别为95.8%和93.3%,而应用基于迭代阈值的最优波长检测方法在753.5 nm波长处检测率高达96.7%。说明利用基于迭代阈值的最优波长检测方法对水稻螟虫的检测更加精确,也说明所获取的特征波段和最优波段为以后水稻螟虫虫害的多光谱成像技术提供了理论参考。

To be able to forewarn rice borers and control the spraying pesticide dosage,to realize the nondestructive detection of rice borers’damage.A feature band detection method based on principal component analysis and an optimal band detection method based on iterative threshold is proposed,the characteristic band and the optimal band of rice stem borers detection are determined,and the images of single band and the combined band are extracted to segment wormholes,to realize the accurate nondestructive detection of rice borers.Firstly,the reflectance information of 120 samples obtained by hyperspectral analysis determined that the spectral region was 450~1000 nm.Band detection method based on principal component analysis characteristics,principal component analysis in the hyperspectral image,in which the first five principal components determine the third principal component images as the best image comparison,and then according to the third principal component in the image,the contribution rate of each band features to select wavelength(668.8 and 750 nm).Finally,global threshold segmentation and image masking are combined to distinguish the wormhole region.Moreover,utilization based on iterative threshold detection method,the optimal band in the visible band 450~750 nm range and near-infrared band 750~1000 nm range application to pick the best single band,mixing distance by single band combination,a combination of single band and band to iterative threshold segmentation.Among them,753.2 nm single band has the best segmentation effect,and 753.5 nm single band is determined as the optimal band.And then extract the band images using a wormhole extraction method based on iterative threshold and morphological processing.Finally,we can distinguish the rice stalk foraminifera region to realize the existence of rice stems infested with borers.The results showed that the detection rates of 60 pest-rice stalks and 60 normal rice stalks were 95.8%and 93.3%respectively,at 668.8 and 750 nm bands by using the principal component analysis-based characteristic band detection method.The optimal band detection method based on the iterative threshold has a detection rate of 96.7%at 753.5 nm band.This indicates that the optimal band detection method based on the iterative threshold is more accurate for the detection of rice borer and also indicates that the acquired characteristic band and optimal band provide theoretical reference for the future multi-spectral imaging technology of rice borers’damage.