A Remote Sensing Identification Method For Jasmine Planting Areas Integrating Phenological And Textural Features

Jasmine as a kind of economic crop,the spatial distribution of jasmine planting area is important for the reasonable planning of jasmine industry layout,promoting the development of jasmine industry and rural revitalization.With the continuous development of remote sensing technology in recent years,satellite image data with the advantages of wide area coverage,timeliness,low acquisition cost and wide data sources have been widely used in the field of agriculture,the countryside and farmers.The main production areas of jasmine cultivation in China are mainly located in the mountainous and hilly areas in the middle and low latitudes,which are susceptible to the influence of cloud and rain climate factors,making it difficult to extract information from jasmine cultivation areas.How to give full play to the advantages between different remote sensing image datasets to realize the extraction of remote sensing identification of jasmine planting areas has become the main direction of current research efforts.Therefore,this paper takes Hengzhou city of Guangxi as the research area,combines the cloud-free remote sensing image data of Landsat 8 and MODIS satellites in2020,uses the spatio-temporal data fusion technology ESTARFM model to reconstruct the annual NDVI time series remote sensing data,then extracts the typical vegetation phenology parameters of Hengzhou city by TIMESAT software,and uses the gray scale coevolution matrix(GLCM)to mine the annual The geometric spatial information and texture features of the satellite data images were used to extract texture features of jasmine planting areas in Hengzhou City.Finally,we combined various machine learning methods and designed different classification schemes for extracting jasmine planting areas in the study area by combining time-series NDVI data,weather parameters and texture features,and analyzed and evaluated the extraction results by combining confusion matrix and statistical information data to investigate the feasibility of extracting low and medium scale multispectral images in jasmine planting areas,and to provide a basis and technical support for promoting jasmine industry layout planning.The conclusions obtained from the study are as follows.(1)The annual NDVI time-series data reconstruction was carried out based on the spatio-temporal data fusion algorithm ESTARFM,and the reconstructed ESTARFM-NDVI data were compared and analyzed with the real Landsat-NDVI data,and the results showed that the accuracy of the reconstructed ESTARFM-NDVI data fusion results reached more than 0.8,which can effectively reflect the NDVI of surface vegetation cover.Three filtering methods,Savitzky-Golay filter,least squares-based asymmetric Gaussian function,and dual logistic function filtering method,were used to reconstruct and fit the NDVI time-series data,and the fitting results showed that the correlation coefficient R of the S-G filter reached 0.927 and the root mean square error was 0.0407,compared with the other two fitting methods,the correlation coefficient R and the root mean square error RMSE were more excellent.The TIMESAT software was used to extract the weather parameters and obtain nine weather parameter features in the study area,and the sample data collected in the field and Google Earth dataset were combined with the weather parameters to analyze the study,and the results showed that the weather parameter information of jasmine in the study area is significantly different from that of other crops,and the construction of annual weather parameter features has obvious advantages for extracting the information of jasmine planting area.The results showed that the annual weather parameters of jasmine in the study area were different from those of other crops.(2)The annual images obtained by Google Earth Engine platform were used to analyze and test the main feature types in the study area,to obtain the degree of separation between different feature types in each band of remote sensing images,to re-elect the band features with better separation to obtain the image texture information,and finally to select four bands of variance and mean parameters,Band 4,5,and 7 texture features of contrast parameters,and the non-similarity of texture information.The results show that the reselected band analysis of feature texture information takes into account the spatial geometric characteristics of different feature types,and alleviates the phenomenon of "same-spectrum and different-spectrum" generated by vegetation spectral index in the process of feature identification.It can effectively improve the performance and accuracy of feature classification in the study area.(3)Based on the reconstructed NDVI time series data,weathering parameters and texture features,three common image classification algorithms were selected for comparison and analysis,and finally three classification schemes designed with the participation of random forest classification algorithm were selected to investigate the optimal classification results of jasmine planting area extraction.The study shows that the weather parameters can correct the misclassification and omission caused by NDVI time series data in the classification process of jasmine;the texture features can effectively identify and distinguish the ungulate features,and eliminate the phenomenon of pores and holes in the extraction process,so as to improve the classification accuracy.The information of jasmine planting areas extracted by the random forest classification method combining NDVI time series data + weather parameters + texture features can better reflect the spatial distribution characteristics of jasmine planting areas in Hengzhou City.