Remote Sensing Identification And Classification Of Typical Crops In Jiangsu Provinc

Crop identification and classification is an important task in agriculture and land management.Accurate crop classification provides valuable information for crop monitoring,yield estimation,crop mapping and crop management.Remote sensing images can capture the spectral,spatial and temporal characteristics of crops and provide valuable information for crop classification,which has now become an effective tool for crop classification.However,the rainy weather and frequent cloud cover in southern China hinder the acquisition of high-quality remote sensing satellite images.In this study,we used Google Earth Engine as the data processing platform and collaboratively used Sentinel-1 and Sentinel-2 satellite data to complete the identification and classification study of three typical crops(rice,wheat and maize)in Jiangsu Province by combining the use of time series analysis,image fusion and machine learning classification techniques.The main research contents and conclusions of this thesis are as follows.(1)A weathering analysis method based on time-series vegetation index.A method based on time series analysis of key phenological periods is proposed to reconstruct vegetation indices using SG filter and apply GLCM method to extract SAR image feature texture.This method can reduce noise,enhance potential signal and characterize crop phenology more reliably.(2)Research on satellite image fusion algorithm based on optical and microwave remote sensing.A fusion algorithm of visible and SAR is proposed,and the link strength of the original PCNN is adjusted according to the entropy of the input image and the fusion strategy using an improved PCNN method to obtain a more balanced and adaptive fusion effect and further improve the texture detail information in the fused image.The proposed method achieves satisfactory fusion results in narrow plots of crops,ridges between agricultural fields,and boundary lines between studied crops and other features.(3)Multi-method typical crop extraction based on cloud platform.The performance of three classifiers,random forest,support vector machine,and classification and regression tree,in classifying rice,wheat,and corn crops was compared using the Jeffries-Matusita method to determine the separability of samples.The experimental results show that the highest accuracy and lowest omission error rate were achieved in the classification study of the three crops when both the NSST-IHS-PCNN fusion algorithm and the random forest classification method were utilized,and finally the area of the crops was accurately mapped in combination with the fusion algorithm proposed in this paper.In this paper,time series reconstruction technique,fusion algorithm of visible and SAR images,and machine learning classification method are used to identify and extract the planting area of typical crops in Jiangsu Province,and the classification accuracy of rice,wheat and corn is 95.25%,96.17% and 91.39%,respectively.The method in this paper shows superior performance and reliable classification results in capturing the fine-scale features of agricultural land cover.