基于偏振纹理特征的枫叶营养诊断方法研究  

Study on Nutritional Diagnosis Method of Maple Leaf Based on Polarization Texture Feature

作　　者：王晓[1] 刘木华[1] 张扬[2] 万奇 刘林林 徐将[1] WANG Xiao;LIU Muhua;ZHANG Yang;WAN Qi;LIU Linlin;XU Jiang(College of Engineering,Jiangxi Agricultural University,Nanchang 330045,China;College of Forestry,Jiangxi Agricultural University,Nanchang 330045,China)

机构地区：[1]江西农业大学工学院,江西南昌330045 [2]江西农业大学林学院,江西南昌330045

出　　处：《江西农业大学学报》2023年第1期202-209,共8页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家自然科学基金项目(32160417,31960363)。

摘　　要：【目的】枫叶的纹理特征可以反映出其营养元素缺失的症状。偏振成像技术是一种增强物体特征对比度的有效方法,且获取过程简便。基于偏振图像的纹理特征模型,对相同生长环境下的4种不同色差枫叶开展了缺素诊断研究。【方法】以灰度共生矩阵的特征为研究依据,分别推导出了能量、熵、反差、相关性等纹理特征作为目标识别参量,构建了基于图像纹理特征的营养诊断模型。通过控制入射光的偏振态,在相同的试验环境下获取了36组不同振动方向上的偏振图像,并基于营养诊断模型提取了枫叶的纹理特征。利用Harris角点检测的特征信号提取算法分别对偏振图像和非偏振图像进行测量。根据检测的结果,分别选取了4块包含样品峰值信号且没有相互干扰的特征信号范围,减少了特征信号提取过程中的误差。结合BP神经网络对特征区域进行预处理,得到每一幅采集图像的不变矩。将枫叶的特征信号作为测试样本,利用MATLAB软件获得诊断模型的残差值分别为:e_(红枫叶)=0.078 3,e_(褐枫叶)=0.101 4,e_(黄枫叶)=0.000 1,e_(绿枫叶)=0.015 3。【结果】在无约束光照条件下,样品图像中纹理的非均匀程度或复杂程度都不会随着偏振片旋转角度的改变而呈现较大的差异。随着偏振角度的增大,偏振图像的4种特征参量都呈现出了明显的变化规律。与非偏振图像相比,偏振图像中各条特征参量曲线之间会随着叶片颜色的加深呈现梯度变化的趋势,且间隔保持稳定。【结论】在偏振散射特性作用下,利用纹理特征可以轻松地对枫叶营养状况进行诊断,且测量精度较高。该模型可以为无人机遥感技术实现树木病虫害的防治提供参考方法,在同一视场中快速实现多目标的识别与诊断。[Objective]The texture characteristics of maple leaves can reflect the symptoms of nutrient deficiency.Polarization imaging is an easy-access and effective method to enhance the contrast of object features.Based on the texture feature model of polarized images,the research of element deficiency diagnosis for maple leaves with four different color differences in the same growth environment was condcuted.[Method]Based on the characteristics of the gray level co-occurrence matrix,the texture features such as energy,entropy,contrast and correlation were derived as the target recognition parameters,and a nutritional diagnosis model based on image texture features was constructed.By controlling the polarization state of incident light,36 groups of polarization images in different vibration directions were obtained in the same experimental environment,and the texture features of maple leaves were extracted based on the nutritional diagnosis model.The feature signal extraction algorithm based on Harris corner detection was used to measure polarized and unpolarized images respectively.According to the detection results,four feature signal ranges containing the peak signal of the sample and without mutual interference were selected to reduce the error in the process of feature signal extraction.Combining BPNN to preprocess the feature area,the invariant moments of each image were obtained.The characteristic signals of maple leaves were taken as the test sample,and the residual value of the diagnostic model obtained by using MATLAB software were e_(Red)=0.078 3,e_(Brown)=0.101 4,e_(Yellow)=0.000 1,e_(Green)=0.015 3.[Result]Under the unconstrained illumination condition,the non-uniformity or complexity of the texture in the sample image will not be significantly different with the change of the rotation angle of polarizer.With the increase of polarization angle,the four characteristic parameters of polarization image show obvious change rules.Compared with the unpolarized image,the characteristic parameter curves in the pola

关 键 词：偏振图像 纹理特征 诊断模型 神经网络 

分 类 号：S718.3[农业科学—林学]