Research On Crop Distribution Information Extraction Method Based On UAV Remote Sensing Technology

In order to explore the potential and applicability of the UAV remote sensing classification method for extracting complex planting structure,this study starts with the extraction method of crop distribution information from UAV remote sensing images,and compares and analyzes the performance of different classification methods for extracting crop distribution information.First,five research areas were selected through ground surveys in Wuyuan County,Inner Mongolia,and their crop types were determined through ground surveys.Among them,study area 1 and study area 2 both have 4 types of crops,study area 3 has 3 types of crops,study area 4 has 5types of crops,and study area 5 has 8 types of crops.Then,based on the UAV remote sensing visible light remote sensing platform and the UAV multi-spectral remote sensing platform,the visible light images of study area 1 and study area 2 and the multi-spectral images of study area 3,study area 4 and study area 5 were obtained,and the images were Perform pretreatment.Finally,based on different remote sensing classification methods,the extraction of crop distribution information in 5 research areas was realized.The main results of this paper are as follows:(1)Clarified the application potential of UAV remote sensing technology in fine classification of crops in small farming areas.First,based on the visible light images of study area 1 and study area 2,pixel-based feature parameter hierarchical classification(PB-HCM)and object-oriented feature parameter hierarchical classification(OB-HCM)are used to study area 1 and Area 2 is used for classification,and the classification accuracy of the two classification methods is above 88%;secondly,based on the multi-spectral images of study area 3,study area 4,and study area 5,the object-oriented support vector machine method(OB-SVM)is used.)And the object-oriented random forest method(OB-RF)to classify study area 3,study area4,and study area 5.The overall classification accuracy is not less than 97.21%.It shows that both UAV visible light remote sensing technology and multispectral remote sensing technology have broad application potential for fine classification of farmland features under highly complex planting structures.(2)Comparative analysis of the performance of pixel-based and object-oriented image analysis methods in crop classification and extraction.Based on the visible light images of study area 1 and study area 2,pixel-based feature parameter hierarchical classification(PB-HCM)and object-oriented feature parameter hierarchical classification(OB-HCM)are used to classify farmland features.The results show that whether it is study area 1 or study area 2,the classification effect of OB-HCM is better than that of PB-HCM.The best extraction methods for bare land,sunflowers,saplings,corn and wheat are all object-oriented methods.(3)Constructed a new vegetation index TVI.The spectral and texture information of five ground objects of bare land,sunflowers,saplings,corn and wheat were analyzed using box plot statistics,and a sapling vegetation index(TVI)based on visible light images was constructed as the feature parameter extracted by saplings.TVI combines pixel-based feature parameter hierarchical classification(PB-HCM)and object-oriented feature parameter hierarchical classification(OB-HCM)to extract saplings in study area 1 and study area 2,with an average extraction accuracy of 92.89 %,indicating that the extraction of saplings based on the TVI index is feasible.(4)The performance of object-oriented random forest classification model(OB-RF)and object-oriented support vector machine classification model(OB-SVM)in crop classification and extraction is compared and analyzed.Based on OB-RF and OB-SVM,the research area 3,the research area 4 and the research area 5 with different planting structures are classified.In the research area 3 with 3 crops,the OB-RF model and the OB-SVM model The classification accuracy is 97.09% and99.13%,respectively.In the study area 4 with 5 features,the accuracy is 92.61% and99.08%,and in the study area 5 with 8 crops,the accuracy is 88.99% and 97.21%,respectively.As the complexity of the planting structure changes from low to high,the accuracy of the two models has decreased.The accuracy of the OB-RF model has dropped by 8.1%,and the OB-SVM model has only dropped by 1.92%.The overall classification accuracy of the OB-SVM model can still reach 97.21% when the plot is finely fragmented and the planting structure is complex.Therefore,compared with OB-RF,OB-SVM has higher classification accuracy when the plot is finely divided and the planting structure is complex,and it is more suitable for fine classification of farmland features with complex agricultural planting patterns.