【摘要】 针对目前高分二号(GF-2)卫星遥感数据在农业领域应用较少,尤其是在农作物识别方面应用缺乏的现象,以GF-2 4 m多光谱遥感影像为数据源,在河南省北部小麦主要种植区域濮阳县,采用监督分类方法(包括支持向量机、人工神经网络和最大似然法)进行小麦种植空间分布信息的快速提取和精度分析。结果表明,3种分类方法对小麦的识别结果非常相似,生产者精度均在96%以上,以支持向量机法最高;用户精度均在98%以上,以最大似然法最高; Kappa系数三者比较接近,均在0. 80以上;总体精度均在82%以上,以最大似然法最高,达85. 15%;错分误差在2%以下,漏分误差在3%左右,对地物的识别误差总体以最大似然法最低,尤其对小麦、水体、光伏电站的识别精度非常高。综合考虑,在采用GF-2进行小麦识别时,建议采用最大似然法。更多还原

【Abstract】 At present,application of GF-2 remote sensing imagery data in agriculture is less,especially in crop recognition. In this regard,the multispectral remote sensing image of 4 m resolution was used to rapidly extract spatial distribution of wheat using methods of support vector machine(SVM),artificial neural network(ANN) and maximum likelihood(MLC) in Puyang county which was major planting area of wheat in the northern part of Henan,and the recognition accuracy was analyzed. The results showed that the wheat recognition results of three classification methods were very similar,producer’s accuracies were all above 96%,which of SVM was the highest; user’s accuracies were all above 98%,which of MLC was the highest; Kappa coefficients were close,all larger than 0. 8; overall accuracies were all larger than82%,which of MLC was the highest with 85. 15%; misclassification errors were less than 2%,and the omission errors were about 3%,overall,MLC method had the lowest recognition error,especially for wheat,water body,photovoltaic power station. Comprehensive consideration,when the wheat was recognized using GF-2,MLC method was recommended.更多还原