Crop Planting Structure Extraction In Luohe Basin Based On Remote Sensing Technology

Planting structure is a combination of spatial distribution and planting pattern information of crops.The investigation of planting structure is of great significance to ensure food security and sustainable development of agriculture.The characteristics of satellite remote sensing,such as large-scale,multi-source,multi-temporal and multi-space,provide important technical support for quick and accurate extraction of planting structure.In recent years,the increasing number of autonomous high-resolution satellites in China provides rich data sources for planting structure extraction.However,there are still some difficulties in using high-resolution remote sensing data to extract the planting structure: first,some data are affected by clouds and hazes,which make it difficult to form high-quality data sets with good time series;second,the applicability of most of the current planting structure extraction methods to high-resolution data needs to be verified;third,most of the crop extraction models need samples as a priori knowledge,at present,the lack of samples is difficult to meet the needs of agricultural monitoring.Based on these situations,in this paper,with the support of MODIS and GF-1 data,the Hutuo River Basin is taken as the research area,the NDVI change rules of crops growth period in the basin is systematically studied,and the effect of decision tree model constructed under the condition of limited GF-1 data on crop recognition and the application potential of generalized DEM method in fast recognition of crops are explored.The main research work and achievements are as follows:(1)By comparing and analyzing the characteristics of Crop Phenology and the NDVI time series curve filtered by Savitzky-Golay method,the change rule of NDVI in each phenological stage of the main crops in the study area was obtained.Based on the results of NDVI time series analysis,the decision tree is constructed by using the time series data set composed of MODIS NDVI data of 23 images in 2018 and NDVI statistical eigenvalues of crop samples,and the planting structure of crops in the basin is extracted.The spatial distribution of crop planting in the region was preliminarily verified by the extraction results.(2)Combined with the six GF-1 images of crop growth period in the study area,from the particularity and difference of NDVI changes in crop growth period,the reasonable threshold and screening rules were studied,and a decision tree wasconstructed to extract the planting structure.The accuracy of obfuscation matrix is evaluated by the samples of visual interpretation of Google Earth's historical 4 m resolution images.The extracting precision of GF-1 decision tree method are more than7% higher than MODIS decision tree method,with an average extraction accuracy of87.99% and kappa coefficient of 0.78.It is proved that the decision tree method based on GF-1 data has a good application effect in the extraction of crop planting structure.(3)As a new crop classification algorithm,the generalized DEM method has the advantages of low data demand and simple operation,which has been proved to have great advantages in winter wheat recognition.In this paper,MODIS and GF-1 data are used to extract a variety of crops in the watershed based on the generalized DEM method.The results show that the generalized DEM method based on MODIS has a general extraction effect,but the average accuracy of the generalized DEM method based on GF-1 is 88.89%,kappa coefficient is 0.80,and the recognition precision is high.The results show that the generalized DEM method is applicable to the extraction of planting structure from various data sources and crops,and it is worth popularizing.