基于多时相GF-1 WFV和高分纹理的制种玉米田识别

Identification Method of Seed Maize Plot Based on Multi-temporal GF-1 WFV and Kompsat-3 Texture

为给监管部门提供更准确的数据,及时发现非法玉米制种区域,根据不同地物在多时相光谱、高空间纹理等特征上的差异,基于163个地面样本、多源时序优选植被指数集和高空间分辨率遥感影像纹理分析的方法,进行制种玉米田识别。通过相关性分析,从GF-1 WFV多光谱影像计算的8个植被指数(VI)中确定6种,多维度反映不同作物光谱差异,并利用随机森林(RF)分类方法实现玉米田块的识别;利用玉米抽雄期的1期0. 7 m Kompsat-3全色影像,构建灰度共生矩阵(GLCM)纹理特征体系,并进行局部二值模式(Uniform-LBP)旋转不变处理,解决了影像中作物种植纹理的方向性问题,同时为体现制种玉米父母本间隔种植的特点,提出了Subtract纹理特征,进一步识别制种玉米田。以新疆维吾尔自治区奇台县为研究区,对本文提出的方法进行实例验证,试验结果表明,制种玉米田识别的制图精度、用户精度分别为93. 34%、99. 19%。

Accurately mastering the planting area and distribution of the seed maize field can provide more accurate data for regulatory authorities,and timely detection of illegal seed production areas.According to the differences of high temporal phase spectrum,high spatial texture and shape of features,the identification of maize seed production field was carried out based on 163 ground samples,multi-source sequential optimization of vegetation index set and texture analysis of high spatial resolution remote sensing images.Through correlation analysis,six vegetation indices(VIs)of the normalized difference vegetation index(NDVI),enhance vegetable index(EVI),normalized difference water index(NDWI),triangle vegetation index(TVI),ratio vegetable index(RVI)and difference vegetation index(DVI)were identified from eight VIs reflecting different growth conditions of vegetation.And the random forest(RF)classification algorithm was used to identify the seed maize field.The gray-level co-occurrence matrix(GLCM)texture feature system was constructed by using the 0.7 m Kompsat-3 image of tasseling stage.It contained five texture features:mean,entropy,contrast,angular second moment(ASM)and homogeneity.At the same time,Subtract texture features were proposed in order to reflect the characteristics of the intercropping of corn parents.Before constructing the texture feature system,local binary patterns(LBP)processing on the image was performed to solve the directional problem of crop planting texture in the image.The random forest was used to identify the seed maize field from maize field classification results.Qitai County in Xinjiang Uyghur Autonomous Region was taken as a research area to verify the proposed method,the results showed that the user s accuracy and mapping accuracy of the seed maize field was 99.19%and 93.34%,respectively.The research result can provide further technical support for the monitoring and supervision of hybrid corn farming in China.