| 摘要: |
| 【目的】采用无人机遥感技术对作物进行分类识别，为及时获取农田信息、制定农田管理策略及产量估测提供技术支持。【方法】采用无人机遥感平台,获取试验区域玉米、桃树、菜花、大豆的可见光正射影像；利用HSV色彩空间转换和纹理滤波,获取不同地物的24项纹理特征与3项色彩特征。分别通过ReliefF算法及基于支持向量机的递归特征消除算法（support vector machine recursive feature elimination,SVM-RFE）进行特征选择与分类，建立6种监督分类模型，利用得到的特征子集对其进行训练，对各模型分类效果进行精度评价。【结果】由SVM-RFE特征子集训练的6种监督分类模型测试集的分类精度均高于80%，分类精度平均提高5.023%，优于ReliefF特征子集训练的监督分类模型，其中SVM-RFE特征子集与支持向量机模型组合对作物的监督分类效果最佳，总体精度达83.417%，Kappa系数为78.60。【结论】基于无人机遥感技术的作物分类识别是可行的。 |
| 关键词: 作物分类识别 无人机遥感 可见光图像 特征选择 监督分类 |
| DOI： |
| 分类号: |
| 基金项目:中央引导地方科技发展专项“精准农业航空技术与装备研发” |
|
| Crop classification based on UAV visible image |
|
LI Zhiming,ZHAO Jing,LAN Yubin,et al
|
| Abstract: |
| 【Objective】The UAV remote sensing technology was used to classify and identify crops,which provides technical support for timely acquisition of farmland information,formulation of farmland management strategies and yield estimation.【Method】The images of experiment sites with maize,peach tree,cauliflower and soybean were obtained by UAV remote sensing platform.Texture filtering and HSV color space conversion were used to obtain 24 texture features and 3 color features of different objects.Feature selection was carried out by ReliefF algorithm and recursive feature elimination algorithm based on support vector machine(SVM-RFE) respectively.Six supervised classification models were trained and established by using the obtained feature subsets.The classification accuracy of each model was evaluated as well.【Result】The classification accuracy of the 6 supervised classification models trained by the SVM-RFE feature subset was higher than 80%,and the accuracy of the classification model was increased in average of 5.023%,which was superior to the classification accuracy of supervised classification models trained by ReliefF feature subsets.The combination of SVM-RFE feature subset and support vector machine model had the best effect on the classification of crops with overall accuracy of 83.417% and Kappa coefficient of 78.60.【Conclusion】The crop classification and recognition based on UAV remote sensing technology was feasible. |
| Key words: crop classification and recognition UAV remote sensing visible image feature selection method supervised classification |
