Sugarcane Extraction Based On Fused HJ Satellite And MODIS Data

Sugarcane is one of the main raw materials of the sugar industry in China,from which Sugar production accounts for more than 90% in China.It plays an important role in the economy of our country agriculture,the output of sugarcane is the second to grain,cotton,and oil.For how to obtain the planting information effectively,timely and accurately,it is one of great significance to formulate relevant import and export plans for the state and relevant departments,stabilize domestic sugar prices and ensure sugar safety,especially for sugar-making enterprises to guide production and scientific planning according to annual planting information.At present,the agricultural sector mainly uses comprehensive statistical to obtain information about the planting area.There is a strong subjectivity in using this method to carry out statistics,which requires a lot of manpower and costs.The accuracy of the statistical data also needs to be verified,and the spatial distribution of sugarcane planting cannot be obtained effectively.Large-scale remote sensing estimation can be made by using remote sensing technology,and the spatial distribution can be accurately obtained.However,due to the shortcoming of satellite return cycle and climate,it is difficult to obtain high resolution images in time series,especially in southern area with cloudy and rainy where sugarcane is concentrated,it will affect the remote sensing of planting area,the accuracy of monitoring is greatly limited.At present,the remote sensing data cannot meet the needs of high resolution at the same time.However,high spatial and temporal resolution remote sensing data can be successfully obtained by using spatial and temporal data fusion algorithm.This study is based on the fusion operation of the domestic environmental satellite(HJ-1A/1B)and MODIS data,and extracts sugarcane crops from southern China,so as to provide the extraction of planting spatial distribution information provides new ideas and methods in fragmented areas of southern China.Based on the MODIS13Q1 and HJ series satellite image data in 2015,and combined with ESTAFRM model,a high spatial and temporal resolution NDVI image data set was constructed.The optimal fitting filtering algorithm was selected to fit and denoise the Time series of NDVI data set,and TIMESA was used to extracts the phenological information of the main crops in the study area.Finally,based on the combination of different bands and random forest algorithm,the remote sensing extraction of large-scale spatial distribution information is realized.The main results are as follows:(1)The correlation coefficients between high spatial-temporal reflectance data and real images based on ESTARFM model are more than 0.8,and the regional correlation coefficients in planting concentration are more than 0.85.The fusion images and real images have overall consistency in spatial distribution.The fusion images can well reflect the spatial details of HJ series satellite images and the temporal variation of MODIS images,and the fused image can reflect the NDVI change of the real vegetation.This algorithm can effectively compensate for the lack of high-resolution images in cloudy and rainy areas of southern China.(2)Using A-G,S-G and D-L filter function algorithms are used to denoise of Time series of NDVI data set space-time fusion and comparing three filter function fitting reconstruction effects.The results show that all three filtering functions can effectively remove some noise of image data set caused by the influence of external environmental factors such as clouds,atmosphere,and sputum.The processed image can better reflect the seasonal variation of ground objects.meanwhile,three kinds of filtering are applied.Compared with the other two filtering algorithms,the S-G algorithm has more advantages and is more suitable for fitting Time series of NDVI data.(3)Based on the fitted NDVI time series dataset and phenological feature dataset,and used the training samples to obtain the phenological feature values and average NDVI values of each object.The results show that there are significant differences in phenological characteristics.in the study area,the maximum annual NDVI of construction land and water is less than 0.5,and the annual NDVI minimum of forest land is greater than 0.6.According to these thresholds,it can effectively remove the Impervious surface,water and woodland in the study area,which can effectively simplify the classification and reduce the redundancy of data.(4)Based on the fitted NDVI time series dataset and phenological feature dataset,and combined with the idea of hierarchical classification,the sugarcane crops were obtained by using the random forest classification.The kappa coefficient was over 85%,and the classification accuracy was high.The results show that it is feasible and feasible to extract sugarcane crops by hierarchical classification.The space-time fusion data can solve the problem of low accuracy of crop classification caused by insufficient temporal resolution of high spatial resolution images in the process of information extraction.(5)According to the comparative analysis of different band combination classification results,the overall accuracy and Kappa coefficient relationship of sugarcane crop image classification in the study area is PH > PH + NDVI > NDVI.The extraction accuracy in the study area is the highest based on phenological data set.The classification results based on a single time series of NDVI data set have obvious errors and omissions,and the classification accuracy is reduced.The results show that the phenological characteristic data set has a greater contribution to the accuracy of sugarcane extraction,and has obvious advantages in extracting sugarcane crops in the south.(6)According to the statistics of the classification results,the sugarcane crops are planted widely in the Zuojiang River Basin of Guangxi.They are mainly planted in Jiangzhou District,Fusui County,Longzhou County,Ningming County,Daxin County of Chongzuo and Shangsi County of Fangchenggang City.Through on-site investigation and comprehensive judgment of Google image,the results of image classification can effectively identify sugarcane space.Compared with statistical data of 8 counties(cities)(regions),sugarcane crop extraction error is below 15% and total error is only 9.27%.The estimated area of remote sensing is basically consistent with the statistical yearbook data.The results show that it can effectively obtain the spatial distribution and area of sugarcane crops based on the fusion of environmental satellites and MODIS data.