基于主成分和云模型的冬小麦种植信息提取研究

在对Sentinel-2卫星遥感影像进行预处理的基础上,利用主成分变化提取小麦主要信息,基于云模型算法开展光谱遥感图像分类。分类时,首先根据训练样本集,由逆向云发生器生成典型小麦的云模型,然后利用云发生器计算出各波段每个象元对小麦地物的平均隶属度,在对各波段的隶属度分析基础上,摒弃含有复杂信息的第1主成分,利用第2主成分和第3主成分信息实现对冬小麦种植空间信息的提取。结果表明,提取小麦种植信息制图精度和用户精度分别为92.78%和99.90%,小麦种植田块的隶属度值因小麦长势和密度的不同有较大的差异,云模型对长势较差、密度较低的小麦像元存在漏分现象。基于云模型的算法精度极高,对小麦地块的识别错分、漏分现象少。该模型有助于冬小麦种植面积的精确提取,对于农业部门进行冬小麦生长监测与产量估测有重要的支撑作用。 

Research on winter wheat planting information extraction based on principal componentand cloud model

The accurate extraction of winter wheat planting area plays an important supporting role in the agricultural sector for winter wheat growth monitoring and yield estimation. In this study, based on the preprocessing of Sentinel-2 satellite remote sensing images, the main information of wheat was extracted by principal component changes, and then a spectral remote sensing image classification method based on cloud model was applied. When classifying, first, according to the training sample set, the cloud model of typical wheat is generated by the reverse cloud generator, and then the cloud generator is used to calculate the average membership degree of each pixel in each band to the wheat ground features. On the basis of analysis, the first principal component containing complex information was discarded, and the second and third principal component information was used to extract the spatial information of winter wheat planting. The results show that the mapping accuracy and user accuracy of extracting wheat planting information are 92.78% and 99.90%, respectively. The membership value of wheat planting fields has great differences due to the difference of wheat growth and density. The cloud model has poor growth and density. The low wheat pixels have the phenomenon of missing points.In general, the algorithm based on the cloud model is highly accurate, and there are few misclassifications and omissions in the identification of wheat field.