基于GF遥感数据纹理分析识别制种玉米

Seed maize identification based on texture analysis of GF remote sensing data

仅利用多时相遥感数据识别作物,其精度难以满足制种玉米识别的实际需求。该文针对制种玉米种植特点,利用国产GF遥感数据,构建了制种玉米遥感识别方法。首先利用多时相国产卫星GF-1 WFV数据,依据研究区作物的物候历,构建各地类EVI时序曲线,提取玉米种植区域;进一步利用抽雄期的GF-2 PAN数据,以田块为对象,通过Sobel边缘检测算子,提取作物纹理信息,并利用Hough变换检测制种玉米田块中的条带状纹理信息,最终提取出制种玉米。该文以新疆维吾尔自治区奇台县坎尔孜乡为研究区,对该文构建的方法进行试验验证,试验结果显示,制种玉米识别精度为90.0%,Kappa系数为0.80。该文不但拓宽了中国国产遥感数据的应用领域,同时也为中国玉米制种监管提供了新的技术支撑。

In order to accurately gather the area and yield information of seed crops and guarantee the seed supplying safety, it＇s necessary to use the remote sensing technology to improve the efficiency and accuracy of the traditional manual statistic survey means. However, the crop identification based on multi-temporal remote sensing data cannot be used in actual seed maize identification because of the low accuracy. Now the domestic remote sensing satellite GF-1 and GF-2 have been launched, and the data characteristic of high spatial and high temporal resolution plays an important role in the field of remote sensing for agricultural condition. With the feature of high spatial and high temporal resolution of the data, this paper builds a method to identify the seed maize focusing on the planting feature of seed maize. At first, according to regional crop calendar, use the domestic GF-1＇s WFV sensor multi-temporal data ranging from April, 2015 to September, 2015 and calculate an EVI（enhanced vegetation index） image for each temporal data which corresponds to every node value at the time series EVI curve, and then extract the maize fields. Since the tassel of the seed maize is removed at the line ratio of 1：6-1：8 so that the female maize＇s spectral reflectance is higher than the male maize＇s, and this feature shows the straight striped texture reflected in the one-meter resolution remote sensing image. Therefore the next stage is regarding the plots as the object and making use of the one-meter spatial resolution remote sensing image of GF-2＇s PAN when removing the tassel to detect the straight texture by the Sobel edge detection operator in the plot object based on the extracted maize area. Then connect the break point at the same direction around a setting appropriate threshold to a line by the Hough transform, and count the number of lines in the plot object. According to whether the sum of lines in the plot is over the setting minimum value, judge whether the crop plot is the seed maize. At last, this paper takes the key seed maize region in Kan＇erzi Town, Qitai County, Xinjiang Uygur Autonomous Region as the test area to verify the above method. The final test result from the confusion matrix shows that the accuracy of classification based on the Hough transform is 90.0%, and the Kappa coefficient is 0.80, which have met the accuracy demands. This seed maize identification method constructed in this paper can obtain the seed maize area with high precision and high efficiency, which not only broadens the application field of domestic remote sensing data, but also supplies a new technology support for the regulation of seed maize. Since the texture of seed maize plot is not always regular but linear within a certain range, it＇s essential to select a reasonable parameter which indicates how much extent of the scatter marked a line in the detecting process of Hough transform. But different crop has different plant regularity so that every crop presents unique texture feature in the high spatial resolution remote sensing image. This method based on the Hough transform is applicable to seed maize identification but limited for other crops, so it＇s the further research to establish crop texture library based on a variety of filtering operators in the field of image process.