基于高光谱成像技术判别马铃薯叶片干旱状态

Evaluation of drought status of potato leavesbased on hyperspectral imaging

为了准确快速地分级评估马铃薯叶片干旱状态,提出了基于高光谱成像技术的马铃薯叶片干旱状态分类方法。于2022年以青薯9号原原种为材料,通过使用高光谱成像仪获取3种不同干旱状态的马铃薯叶片,提取各类样本光谱反射率信息420个,讨论了4种光谱数据预处理方式对建模的影响。使用基于随机森林的交叉验证递归特征消除算法(RF-RFECV)与竞争性自适应重加权抽样法(CARS)进行特征波长选择,结合极端随机树(extremely randomized trees,EXT)构建了马铃薯叶片干旱状态的分类模型。结果表明,本研究构建的3个干旱状态分类模型,其测试集模型精度均高于85%,其中SNV-RF-RFECV-EXT模型表现最佳,测试集预测准确率达92.14%。同时,为直观地显示马铃薯叶片的干旱状态,选用建立的SNV-RF-RFECV-EXT模型对叶片进行干旱程度可视化,通过不同颜色直观显示叶片干旱状态,为马铃薯叶片干旱状态的判别提供了新方法。

To evaluate the drought status of potato leaves accurately and quickly,a hyperspectral imaging method was employed to assess the drought status of potato leaves.In 2022,the original Qingshu No.9 seed was used as the test material.Three types of drought status were obtained using hyperspectral imaging.A total of 420 spectral reflectance measurements from various samples were extracted and the influence of four spectral data preprocessing methods on modeling was discussed.A classification model for the drought status of potato leaves was constructed using the random forest cross-validation recursive feature elimination algorithm(RF-RFECV) and the feature wavelength selection method of competitive adaptive reweighted sampling(CARS) methods combined with extremely randomized trees(EXT).The results showed that the three drought status classification models constructed in this study had accuracy higher than 85% in the test set,among which the SNV-RF-RFECV-EXT model performed the best with a prediction accuracy of 92.14% in the test set.To visually assess the drought status of potato leaves,the SNV-RF-RFECV-EXT model was applied to depict the degree of drought.Different colors were used to represent varying levels of drought,offering a novel approach for distinguishing the drought conditions of potato leaves.