基于信息熵特征选择的小麦冠层叶绿素含量估测方法

Estimation Method of Wheat Canopy Chlorophyll Based on Information Entropy Feature Selection

为运用图像颜色特征估测作物的叶绿素含量,以自然环境下的小麦冠层图像为研究对象,提出一种基于熵权法的颜色特征选择方法,并应用机器学习方法建立小麦冠层叶绿素含量估测模型。熵权法通过信息熵来衡量颜色特征指标权重,实现冠层图像特征排序,机器学习方法选用多元线性回归(Multiple linear regression, MLR)、岭回归(Ridge regression, RR)和支持向量回归模型(Support vector regression, SVR)估测小麦冠层叶绿素含量。试验结果表明,与皮尔逊相关系数法和主成分分析法选取的特征集进行对比,熵权法得到a^(*)、R-G-B、R-G、(a^(*)+b^(*))/L、a^(*)/b^(*)、(R-G)/(R+G+B)、(R-B)/(R+B)、H/S、(R-G)/(R+G)等9个特征组成的特征集,可以利用较少的特征指标达到最优的预测效果。在选取相同特征指标参数的情况下,SVR的预测能力优于其它模型,其R^(2)和RMSE的平均值分别为0.80、1.89,相比于MLR和RR模型R^(2)分别提升2.8%、1.1%,RMSE分别下降0.13和0.05。将基于熵权法建立的SVR模型应用到2021年采集的小麦冠层图像数据,结果表明模型具有很好的稳定性。

Chlorophyll is an important indicator reflecting the nitrogen nutrition status of crops, and its content is closely related to crop growth and development, photosynthesis capacity and crop yield. With the increasing maturity of image processing technology, choosing image color features to estimate the chlorophyll content of crops has become an important technical means. Taking the wheat canopy image in the natural environment as the research object, a color feature selection method was proposed based on the entropy weight method, and machine learning methods were applied to establish a wheat canopy chlorophyll content estimation model. The entropy method used information entropy to measure the weight of color feature indicators to achieve the canopy image feature ranking. The machine learning method used multiple linear regression(MLR), ridge regression(RR) and support vector regression models(SVR) to estimate the chlorophyll content of wheat canopy. The experimental results showed that compared with the feature set selected by the Pearson correlation coefficient method and principal component analysis, the entropy weight method obtained a^(*), R-B-G, R-G,(a^(*)+b^(*))/L, a^(*)/b^(*),(R-G)/(R +G+B),(R-B)/(R+B), H/S,(R-G)/(R+G) and other nine features. The feature sets can use fewer feature indicators to achieve the best prediction effect. In the case of selecting the same characteristic index parameters, the predictive ability of SVR was better than that of other models, and the average values of R^(2) and RMSE were 0.80 and 1.89,compared with MLR and RR models, its R^(2)was improved by 2.8% and 1.1%, RMSE was decreased by 0.13 and 0.05,respectively.The SVR model based on the entropy weight method was applied to the wheat canopy image data collected in 2021, and the results showed that the model had good stability. The above research results showed that image processing technology and machine learning methods had very good application value in the estimation of chlorophyll content of crops, providing an important theoretical basis for image-based estimation of chlorophyll content of field crops.