Oasis Main Food Crops Remote-sensing Recognition And Yield Estimation Based On Multi-temporal GF-1 WFV Images

The traditional method for extracting crop planting area was mainly through large area field measurements or statistical department report step by step, this method not only consumes human, material and financial resources, but aslo lack of crop spatial distribution of information.The development of remote sensing(RS) technology for crop planting area monitoring provides a new mean of statistics, which is able to monitore crops information quickly and efficiently in a large scale. At present, remote sensing technology has been widely applied in agricultural research fields, such as agricultural resources survey, crop planting area, yield estimation, agricultural disaster monitoring and digital agriculture. This not only helps to improve the agricultural scientific decision level of related departments, but also provide better service for agricultural production, which accurately monitor crop spatial distribution information in a large scale in real time.This article selects continental river basin in arid oasis as the study area, based on domestic GF-1 satellite images, after dealing with the S-G filter based GF-1 WFV-NDVI time-series data, by applying hierarchical decision tree method, the research area main food crops of corn and wheat distribution were extracted, using the background investigation of major crops in 2015 of Gansu province food crops in the spatial distribution of data sets to verify this extraction accuracy of corn and wheat. Combining with the production data and corn belt GF-1 WFV-NDVI value relevance analysis, the best yield estimation issue was selected. Then by applying regression analysis method the spring maize yield estimation model of remote sensing was built, then the results can be obtained with using production statistics method.Through research, this paper has summarized the following main conclusions:(1) Based on crops growth/WFV puberty GF-1 satellite images, through phase sequence GF1 Savitzky-Golay build filtering-NDVI dataset, and combined with crop phenological characteristics at the same time, the grain crops in arid areas can be identified and extracted to estimate the spatial distribution of information production.(2) According to the different development stages of the crop in the study area GF-1 WFV-NDVI images, the spring corn and spring wheat planting area can be axtracted by using hierarchical decision tree method for classification. Then compare the area with automatic interpretation results which commonly use SVM(Support Vector Machine) and ANN(Artificial Neural Network) man-machine interactive computation, we can see the accuracy has been improved a lot. The average extraction accuracy reached 93.59%, the average relative error was 6.84%, the mapping accuracy is 95.3%, the user accuracy is 89.2%, achieved the practical application of crop-yield assess by remote sensing.(3) With spring corn in the study area as an example, according to the production data and corn belt GF-1 WFV-NDVI value relevance analysis, the highest correlation coefficient(0.7981) of heading stage was selected as the best yield estimation. With regression analysis method the spring maize yield estimation model of remote sensing was built, then use research statistical data in 10 villages and towns of 2015 to verify the annual yield estimate model calculation results. Production range of relative error between 1.69%-12.25%, the average error of 6.44%.The results show that remote sensing yield estimation model can better for spring maize yield estimation remote sensing.Based on the spatial resolution 16 meters high marks satellite data recognition and production estimates for the basic unit research.The research results show that the crop planting area in a large scale, the extracting method and yield estimation model can reach high precision. But for pure crops like yuan coverage is hard to achieve the desired effect, need to choose higher spatial resolution remote sensing data for a more detailed study.This study summarized a set of data based on the domestic high marks in continental river basin in arid area, the crop distribution and production information are more fast and efficient access to large scale method, which can be applied widely in similar area, has a certain practical significance.