Crop Classification Using Multi-features Of Chinese Gaofen-1/6 Sateliite Remote Sensing Images

Remote sensing technology with quick,easy,macro,lossless and objective features is widely used in the agricultural applications for its,such as agricultural land use,agricultural resources survey,meteorological disaster monitoring and crop condition monitoring,crop yield estimation and so on.Series of satellites of China High-resolution Earth Observation System provide a continuous and reliable data source for agricultural application.Especially,GF-1 and GF-6 satellites combine the characteristic of high spatial resolution,high temporary resolution,wide coverage and multi-spectral,and they can be considered as the important satellite data for the acquisition of crop information.This thesis focuses on features mining of GF-1/6 images for the classification of different crops.Based on the characteristics of high resolution,high temporary resolution and red-edge bands,three parts are considered: 1)object-oriented classification based on high spatial resolution characteristics of crops,2)objectoriented crop classification based on time series data and 3)crop classification based on red edge information.Characteristics of GF-1/6 satellite images are fully exerted to improve the accuracy of crop classification,and provide support for crop planting area monitoring.The spatial resolution of panchromatic data of GF-1/6 is up to 2 meter,which provides rich texture feature for crop identification.Based on the fused data with 2 meter of GF-1 in the study area,10 vegetation indexes related to crop and 16 texture features were extracted,then using the method of random forest to evaluate importance scores for the features,and the relationship between the different number of features and crop classification accuracy was analyzed.Respectively using 5,10,15,20,25 number of features according to the importance,it is found that classification accuracy reach to balance with 15 number features,and the increase of features quantity has little effect to improve classification accuracy.Based on the 15 features crop classification was implemented with object-oriented,and the overall accuracy of classification results is 96.0622% which increases about 5% than the classification result with pixel-based.Results show that the multiple features of object-oriented classification method for high resolution crop classification can obtain good result,and using relevant vegetation index and texture features of near infrared wave band to participate classification,can guarantee the classification accuracy and reduce data redundancy.NDVI time series data can describe the different period of crop growth status,and present the process of the seasonal rhythm changes for crop.Different crops have different phenological phase,and the NDVI curve is also different,therefore,it can be to analysis NDVI curve variation characteristics and trends according to the crop phenological calendar in order to distinguish crop types.GF-1/6 satellite broadband data have high time resolution,and large coverage,and provides continuous and reliable data source for obtaining change information of wide range crops.In the study area,the county crops object-oriented classification was carried out by using the NDVI time series data in entire crop growth cycle from the GF-1 WFV.Results show that using the time-series NDVI in a cover crop growth period of GF-1 WFV,it is able to depict the crops growth condition accurately and distinguish the various types of crops effectively.Using object-oriented classification method with combining NDVI time series data for crop classification,high precision were obtained.GF-6 will add two red edge band which can provide abundant spectral information and more effective means for crop classification.This research use Sentinel-2A satellite data as the similar data source to carried out the impact of red edge band for crop classification.The results show that adding two red edge band can improve the accuracy of crop classification.In the study area,the maximum likelihood method,artificial neural network and support vector machine(SVM)classification method are implemented for crop classification,and the classification accuracy obtained by three classification method with two red edge band are up by more than 14% than without red edge band.Both Red edge band 705 and red edge band 705 can improve the recognition and classification for crops,and the classification accuracy using two red edge bands is superior to the classification accuracy using alone a red edge band.