| Timely and accurate acquisition of regional crop planting structure information is of great significance to agricultural production management and policymaking.UAV remote sensing has the characteristics of high spatial resolution,fast data acquisition,simple operation,and low cost.UAV can quickly collect images of a certain area,and obtain more accurate spatial characteristic information of crops with higher processing efficiency.UAV is of great significance to the development and application of crop monitoring technology and has been widely used in crop classification and area estimation.It has become an important supplement to satellite remote sensing and makes up for the applicable limitation of satellite remote sensing images.In this thesis,high-resolution UAV remote sensing images are used as data sources,and some reclamation areas in Shihezi,Xinjiang are selected as research areas.Firstly,a variety of vegetation indices are analyzed and compared to constructing UAV time-series remote sensing images.Secondly,the spectral features and texture features of UAV images are fused.Finally,we used a variety of classifiers to try to classify crops,and evaluated the final classification results with the help of a variety of indicators such as Overall Accuracy,Kappa coefficient,Commission Error,and Omission Error,to screen out the crop fine classification method suitable for the relatively complex planting structure areas.This study can further improve the accuracy of crop classification,provide new research ideas for UAV remote sensing application and crop classification methods,and provide technical support for the formulation of agricultural sustainable development policies.The main research contents and conclusions of this thesis are as follows:(1)Construction of UAV time-series remote sensing image.To further improve the accuracy and efficiency of crop classification,four commonly used vegetation indices(GI,SRI,NDVI,and NGRDI)were selected as spectral feature space for analysis,and the separability of crop samples under different time series vegetation indices was calculated by Euclidean Distance(ED).The results showed that among SRI indices,cotton,squash,spring wheat and corn had the highest separability among each other.Therefore,the SRI index was finally selected as the preferred feature to construct UAV time-series remote sensing images.In addition to eliminating redundant features,it can effectively solve the disadvantages of a single period image and improve the classification effect of crops.(2)Classifier selection.In this thesis,crop classification is studied from pixel-based and object-oriented levels respectively.It has been verified that RF has the highest Overall Accuracy and Kappa coefficient,reaching 88.94%、0.8534,and 92.07%、 0.8953,respectively.At the same time,for cotton,spring wheat,zucchini,and corn,RF’s Producer Accuracy and User Accuracy are all greater than 92%.This indicates that RF has achieved a good classification effect in crop remote sensing image classification both based on pixel and object-oriented,and has high accuracy and stability for different crop types.(3)Selection of classification methods.Pixel-based crop classification results have a certain "salt and pepper effect" and "mixed pixel" phenomenon,while the field boundary based on object-oriented classification results is more obvious,which is more consistent with the real planting structure space of crops.Compared with pixel-based crop classification results,the Overall Accuracy and Kappa coefficient of RF also improved by 3.13% and 0.0419,respectively.At the same time,through the contrast analysis:using the UAV remote sensing image of crop classification is better than the sentinel’s satellite remote sensing image,the effect of the Overall Accuracy and Kappa coefficient increased by 3.51% and 0.0458 respectively.Thus,using UAV time-series remote sensing images,the object-oriented classification method is more suitable for crop fine classification. |
PDF Full Text Request