基于图像颜色与形状特征的幼龄沉香全氮含量预测

Prediction of Total Nitrogen Content in Young Aquilaria sinensis Based on Image Color and Shape Characteristics

为了促进数字图像处理技术在林木营养诊断中的应用,实现对林木生长状态及养分含量的实时监测,本研究以濒危珍贵树种沉香为对象,构建了3种基于图像颜色特征与形状特征的幼龄沉香全氮含量预测模型,为幼龄林木的营养诊断提供了理论依据。首先,根据边界距离与设定误差的大小确定最佳K值,运用改进的K-Means算法提取前景图像。然后,分离前景图像的R、G、B三通道并分别计算均值,根据图像颜色空间转换公式,将图像分别转换到HIS、Lab颜色空间下,得到色调(H)、饱和度(S)、明度(I)、亮度(L)、红到绿通道(a)、黄到蓝通道(b),共计获得9种颜色特征。寻找前景图像的最小外接矩形,计算前景图像的面积(CA),前景图像最小外接矩形的面积(RA)、周长(RC)以及矩形度(RD),共计获得4种形状特征。最后,分别对颜色特征、形状特征、颜色特征+形状特征进行主成分分析,以获得的3类主成分为自变量构建幼龄沉香全氮含量预测模型,同时对构建的3种模型精度进行检验。结果表明,改善K值选取方式可以降低K-Means聚类分割算法的不确定性,增强算法的分割效率,可以实现对沉香可见光图像的精准分割。本研究构建的3种幼龄沉香全氮含量模型预测能力良好,其中基于单图像参数构建的模型精度基本一致,但基于形状特征构建的模型使用参数更少,相对建模效率更高;双图像参数模型较单图像参数模型的使用参数更多,但拟合度更好、精度更高,在实际应用中可根据不同需要进行选择。本研究运用了不同图像特征构建全氮模型,更好地实现了对幼龄林木全氮含量的无损估测,为精准林业提供了一定的参考。

In order to promote the application of digital image processing technology in forest nutrition diagnosis and realize real-time monitoring of forest growth status and nutrient content,three prediction models of the total nitrogen content of young Aquilaria sinensis Lignum Resinatum based on image color and shape characteristics were constructed in this study,which provided a theoretical basis for nutrition diagnosis of young forest tree.Firstly,the optimal K value is determined according to the boundary distance and the size of the setting error,and the improved K-Means algorithm is used to extract the foreground image.Then,separate the three channels of R,G,and B of the foreground image and calculate the average value respectively.Then,the R,G,and B three channels of the foreground image are separated and the mean values are calculated respectively.According to the image color space conversion formula,the image is con-verted to HIS and Lab color space respectively,and the hue(H),saturation(S),brightness(I),brightness(L),red to green channel(a),yellow to blue channel(b),and a total of 9 color features are obtained.Find the minimum circum-scribed rectangle of the foreground image,calculate the area(CA)of the foreground image,the area(RA),perimeter(RC),and rectangularity(RD)of the minimum circumscribed rectangle of the foreground image,and obtain four shape features in total.Finally,principal component analysis was performed on the color features,shape features,and color features+shape features,and the obtained three types of principal components were used as independent variables to construct a prediction model for the total nitrogen content of young A.sinensis,and the accuracy of the three models constructed was tested.Finally,the principal component analysis of color feature,shape feature,and color feature+shape feature was carried out respectively,and the three principal components obtained were used as independent vari-ables to construct the prediction model of the total nitrogen content of young A.sinensis,and the accuracy of the three models was teste.The results show that improving the K value selection method can reduce the uncertainty of the K-Means clustering segmentation algorithm,enhance the segmentation efficiency of the algorithm,and achieve accurate segmentation of A.sinensis visible light images.The three models of the total nitrogen content of young agarwood con-structed in this study had good prediction ability.The model accuracy based on single image parameters was basically the same,but the model based on shape features used fewer parameters and had higher relative modeling efficiency.The two-image parameter model uses more parameters than the single-image parameter model,but the fitting degree is better and the accuracy is higher.In practical applications,it can be selected according to different needs.In this study,dif-ferent image features were used to build a total nitrogen model,which better realized the non-destructive estimation of the total nitrogen content of young trees,and provided a certain reference for precision forestry.