Study On Identification Irrigated Crop Types In Arid And Semi-arid Regions Based On Time-series NDVI

Water shortages and fragile ecosystems have seriously restricted the development of agriculture in arid and semi-arid regions. Timely and accuratecollection of cropinformation canprovide scientific references forrationalplantation and water-saving irrigation, in order to get the healthy development ofthe agriculture.The remote sensinginformation extraction of crops will get optimal results withits wide coverage and high precision, because NDVI time series imagescan objectively and effectivelyreflect the vegetation green degree and the rate of photosynthesis,as well astheannual and seasonal variation.Theagricultural region on Tianshan Mountain North Slope Economic Development Belt, a typical arid and semi-arid region inXinjiang, isselected as the study area to identify the types of the irrigated crops. The NDVI time series are established based on the Landsat-8 multi spectral images, with the Landsat-7 as the assistant, to classify and identifythe irrigated crop types. The main research contents and conclusions are as follows:(1) Firstly, the filling method oflinear interpolation is used to repair the missing data of the Landsat-7in order to establish the complete time series NDVI. The missing data interpolation is simulated byusing theLandsat-8, which is masked with strips,and gets a result with the overall deviation less than 15%.It shows that the linear interpolation method can improve the quality of the missing data images effectively.(2) Secondly,S-G filtering theory is studied to select the optimal parameters to smooth the reconstructed NDVI time series, with the window size of 5 and the polynomial number of 3.Comparing the curves of NDVI time series before and after filtering can found that S-G filtering method has improved the overall level of NDVI time series. So that it canbe more reliable to reflect the changes of crop growth andthe classification and identification using the NDVI time series will be more effective.(3) Then, the MLC and the SVM are used in the classification tests, and the classification results ofthe same trainsamples with different sizes are compared. The same train samples with different sizes are used. The results indicated that a larger training sample will get a better classification in the tests using MLC, with theaccuracy increased from 75% to 84%.However, in contrast, the SVM classification method obtains consistent classification results, with the overall precisions higher than 85%.(4) Finally,synthetically accuracy statistics of different crops show thatsingle cropswith distinctive regularity can be distinguished with a very high accuracy, and a variety of crops with similar regularity are be distinguished with slightly lower accuracy.To conclude, the classification tests get an overall accuracy higher than 86%of all crops, and the macro pattern of crop is effectively reflected.