Abstract:

In order to improve the pertinence of practice in a large scale farm， and explore an efficient method for delineating management zones， a 30.8 hm² typical black farmland of spring maize in Heilongjiang province was taking as object to study management zones. Based on the multispectral images at silking stage of maize obtained by unmanned aerial vehicles（UVA） equipped with Sequoia sensor， a method of multiresolution segmentation combined with fuzzy-C means clustering algorithm was used to delineate the farmland. At the same time， segmentation by fuzzy-C means clustering algorithm based on grid sampling and soil properties， including soil organic matter， available NPK， soil volumetric moisture content， soil conductivity， and pH， was conducted as a contrast. The coefficient of variable （CV） of soil properties and yield were chosen to evaluate effect of the 2 methods. The results showed that the partition based on UAV images was delineated to 4 optimal management zones named as M1， M2， M3， M4， and their yields were 7 597.53，8 236.35，8 686.98，9 119.93 kg·hm^-2， respectively. There were significant differences in maize yield between each zones， and in soil nutrient between M1， M2， M3 zones， and the CV of soil nutrient， crop yield among each zones were much lower than that before the management zones. It was also delineated to 4 optimal zones determined based on soil properties named as S1， S2， S3 and S4， and their yields were 7 754.81，8 173.44，8 860.05，9 153.23 kg·hm^-2. There were significant differences in soil nutrients and soil water content between each zones （P<0.05）， and the homogeneity of soil properties within the zones was improved. The 2 methods had certain similarity in spatial distribution. The spatial coincidence degree under the same level zones were 40.00%， 46.51%， 57.45%， 59.38%， respectively， and the overall coincidence degree was 51.32%. Further，the results based on UAV multispectral images provided basis for farmland management zoning in the absence of soil data.

Key words: management zones, multispectral, UAV, crop growth, soil nutrients

摘要：

为了提高大尺度农田管理的针对性，探寻低成本高效的分区方法，以黑龙江省典型黑土区30.8 hm²春玉米田为对象进行分区研究。基于吐丝期无人机多光谱影像，使用多尺度分割与模糊聚类相结合的方法进行分区，同时基于播种前的土壤养分（土壤有机质、速效氮磷钾）、土壤体积含水率、电导率、pH进行模糊聚类分区并作为对照，对分区间春玉米产量和土壤养分进行方差分析，并对分区内变异系数进行比较以评价分区效果。结果表明，基于无人机影像得到4个较优管理分区为M1、M2、M3、M4，各分区产量分别为7 597.53、8 236.35、8 686.98、9 119.93 kg·hm^-2，各分区间产量差异显著，其中M1、M2、M3间土壤养分差异显著（P<0.05），分区内作物产量和土壤养分的变异系数降低；基于土壤数据确定4个分区，即S1、S2、S3、S4，春玉米产量分别为7 754.81、8 173.44、8 860.05、9 153.23 kg·hm^-2，分区间土壤养分、土壤水分差异显著（P<0.05），分区内部土壤均一性提高。综合来看，2种方法的划分结果在空间分布上具有一定的相似性，同级分区的空间重合度分别为40.00%、46.51%、57.45%、59.38%，整体重合度为51.32%。当缺乏土壤数据时，无人机多光谱影像可为农田管理分区提供参考依据。

关键词: 管理分区, 多光谱影像, 无人机, 作物长势, 土壤养分