Study On Crop Area Extraction Using Remote Sensing Technology Based On Unmanned Aerial Vehicle

The traditional statistical methods are subjective to large-scale,large-scale crop monitoring and identification.The errors are large and the crop information obtained has great limitations.Using drone remote sensing image technology to extract crop planting information can quickly estimate crop planting area.Firstly,this paper uses UAV remote sensing experiment to obtain multiple images with overlapping regions.Using Agisoft photoscan software to splicing and reconstructing the complete image of the experimental area,using the multi-scale segmentation method to divide the experimental region into several objects,and based on statistical methods to extract the objects.Spectral features,geometric features,and texture features;then,a binary logistic regression model for identifying paddy plots was established.The feature indices were shape index,red mean value,red standard deviation,maximum difference measure,and gray level co-occurrence matrix homogeneity.Degree cooccurrence matrix non-similarity.The results show that the accuracy rate of the model identification training sample set is 100%,the correct rate of the identification test sample is 97%,and the model is applied to identify and verify the rice paddy field.The overall accuracy rate is 98%.Finally,based on the cumulative pixel method,the area of paddy fields was calculated and compared with the results of visual interpretation.The area error was less than 3.5%.The method used to identify paddy fields was effective and the accuracy of area measurement was high.Therefore,this research has certain applicability to the use of UAV remote sensing images to survey rice cultivation information.The main research contents and conclusions are as follows:(1)Data acquisition and preprocessing.The six-rotor UAV equipped with camera sensors was used for data acquisition and was preprocessed with Agisoft PhotoScan software.(2)Object-oriented analysis.Using feature acquisition statistics,segmentation methods and classification algorithms,and statistics,the establishment of two-class Logistic regression model for the identification of rice fields,making the object-oriented classification accuracy.The extraction accuracy is 97%,and the area error is within 3.5%.(3)Analysis based on pixel method.This paper introduces the maximum likelihood method under the supervision classification block and the K-means classification under the unsupervised classification,and uses these two methods to classify the remote sensing images in the test area and the verification area.The extraction accuracy is not ideal,and the area error reaches 17%.(4)Accuracy evaluation.The area-accuracy comparison evaluation was performed on object-oriented classification based on binary classification and pixel-based maximum likelihood.