Object-oriented Multi-temporal Alfalfa Extraction Method

Alfalfa has the characteristics of easy survival,high yield,good quality and can effectively maintain the stable and healthy development of agriculture and livestock,so it is of great significance to obtain the distribution data of alfalfa.The popularization of remote sensing technology makes the extraction of alfalfa more efficient and accurate.The combination of two or more methods has been widely used in crop recognition,which can integrate the advantages of multiple methods and improve the recognition performance.In this study,Alukorqin Banner and Zhangye City were taken as the research areas.With the help of multi-timeseries remote sensing images obtained from the medium-resolution sensor GF1WFV,An object-oriented approach based on threshold segmentation is proposed to provide technical support for alfalfa extraction.This approach also can provide professional technical support for the manufacture and production of alfalfa,and can guide practice,operation management and decision-making judgment.The research in this paper is as follows.(1)After analyzing the relationship between the cycle law of alfalfa and environmental conditions,it is found that the spectral values of remote sensing images change with the number of alfalfa mowings per year.Complete information on the distribution of alfalfa can be obtained by studying images during the growing period of alfalfa.(2)In this study,the GF1-WFV data set with relatively high temporal and spatial resolution was selected to extract the spatial distribution information of alfalfa,and an object-oriented LDVI(Linear Difference Vegetation Index)alfalfa extraction method was proposed.In detail,the LDVI time series are built after performing object-oriented segmentation on each time-series image.Then,the difference in spectral reflectance characteristics between alfalfa and other features is enhanced by setting rules so that all kinds of interfering ground object can be eliminated by thresholds.Finally,this method enables the automatic acquisition of the spatial distribution of alfalfa,which can improve the problem of incomplete patch extraction and mixed pixels.(3)The effects of different time densities on alfalfa identification compared through.The results show that the denser the time density of remote sensing images obtained during the growing period of alfalfa and the more images are,the higher the accuracy of the object-oriented LDVI-based alfalfa identification method.The accuracy of the proposed method was verified by comparing with other traditional methods.It was found that the proposed method had the highest accuracy in alfalfa extraction,reaching 97.3%and 95.2%in Alukhorqin Banner and Zhangye City.In contrast,the accuracy of supervised classification,unsupervised classification and threshold segmentation methods were poor and unstable in different enviroments.In conclusion,the proposed method not only expands the connotation of remote sensing ground object classification method,but also obviously improves the problem of low u niversality caused by environmental and climatic factors.Besides,it also has the outstanding characteristics of flexible application,automation and broad development prospect.