高级检索+

基于多核主动学习和多源数据融合的农田塑料覆被分类

Classification of Agricultural Plastic Cover Based on Multi-kernel Active Learning and Multi-source Data Fusion

  • 摘要: 通过引入多源多时相卫星遥感数据,提出了一种基于多核主动学习的农田塑料覆被分类算法,实现农业塑料大棚和地膜的精准分类。首先基于多时相Sentinel-1雷达和Sentinel-2光学遥感影像,提取其光谱特征、纹理特征等,以构建多维特征空间。然后构建多核学习模型,实现多源、多时相特征的自适应融合。最后构建基于池的主动学习策略,通过引入训练样本的淘汰机制,进一步提升分类模型的泛化能力。试验结果表明,本文所提分类方法的总体精度为95.6%,Kappa系数为0.922,相较经典支持向量机、随机森林、K近邻、决策树、AdaBoost模型,多核学习模型精度提高5.7、12.1、11.4、22.3、10.3个百分点;且在相同分类精度下,主动学习较被动学习可减少一半以上的标签数据;同时相较仅使用单时相及单传感器遥感影像而言,精度分别提高3.7、12.7个百分点。结果表明,多核主动学习能够有效进行多传感器、多时相数据融合,并可以在小样本条件下取得更高的分类精度,从而为农田塑料覆被的遥感监测提供模型参考。

     

    Abstract: An agricultural plastic covering classification algorithm was proposed based on multi-kernel active learning to achieve accurate classification of agricultural greenhouses and mulch film by introducing multi-source and multi-temporal satellite remote sensing data, and their spectral features and texture features were firstly extracted to construct a multi-dimensional feature space based on the multi-temporal Sentinel-1 radar and Sentinel-2 optical remote sensing data. And then, a multi-kernel learning model was constructed to realize the adaptive fusion of multi-source and multi-temporal features. Finally, a pool-based active learning strategy is constructed to further improve the generalization ability of the classification model by introducing an elimination mechanism for training samples. The test results showed that the overall accuracy of the proposed classification method was 95.6%, the Kappa coefficient was 0.922. Compared with that of the classic SVM, random forest, KNN, decision tree, AdaBoost model, the accuracy of the active learning model was improved by 5.7, 12.1, 11.4, 22.3 and 10.3 percentage points. And under the same classification accuracy, active learning can reduce more than half of the label data than passive learning. The accuracy was improved by 3.7 and 12.7 percentage points, respectively, compared with using only single-phase and single-sensor remote sensing images. The research results showed that multi-kernel active learning can effectively perform multi-sensor and multi-temporal data fusion, and can achieve high classification accuracy under small sample conditions. It can provide model reference for remote sensing monitoring of agricultural plastic cover.

     

/

返回文章
返回