The Research Of Rape Planting Area Measurement Technology Based On High Resolution Satellite Remote Sensing

Oil seed rape is one of the four major edible oil crops in China.Timely and accurate acquisition of oil seed rape spatial distribution is of great significance for the adjustment of planting structure,food security,and national economic and social development.However,the traditional field-survey based methods are difficult to meet the requirements of regular monitoring of oil seed rape spatial distribution in a large scale of ranges.Alternatively,the successful launch of Gaofen-series satellites since 2013 in China has provided important data support for rapid and accurate mapping of crop spatial distribution and area estimation.In this study,we aimed to extract the oil seed rape and estimate its area rapidly and accurately using Gaofen-2(GF-2)satellite images.Some villages in Changde,which are located in the rape production belt in the Yangtze River Basin of China,were selected as the studying area.The main works and studies we did are summarized as follows:(1)Images preprocessing and rape samples preparing: Firstly,GF-2 images were preprocessed,including Radiometric calibration,orthorectification,image registration,atmospheric correction and image fusion,to get the reflectance images,which can present the true information of ground objects.Secondly,the rape pixels were labeled and sample pairs were conducted with the using of GIS software.(2)Rape extraction experiments based on Deep Learning Model:The deep semantic segmentation model,Deep Labv3+,was used to extract the oil-seed rape.This model has the abilities of multi-scale feature learning using Atrous Spatial Pyramid Pooling(ASPP),edge information extraction with structure of Encode-Decode,and fast computation by applying depthwise separable convolution,that makes it possible to extract the rape rapidly and accurately.(3)Rape extraction experiments based on object-oriented algorithm: The FNEA＿KNN object-oriented algorithm,which is the combination of FNEA and KNN algorithm,were adopted to extract the rape.This algorithm includes four main steps:image segmentation using FNEA algorithm,feature selection,rule set establishment,rape extraction using KNN algorithm,and accuracy validation.(4)Accuracy evaluation and comparison between two models: Four different accuracy indexes including precision(P),recall(R),F1-score and overall accuracy(OA)were used to evaluate the extraction accuracy of rape in four selected typical villages.The results showed that compared with FNEA＿KNN,Deep Labv3+ got higher values in all these four indexes,with P = 95.66%,R = 93.46%,F1-score = 94.54%,and OA =99.69%,indicating that Deep Labv3+ is more suitable for rape extraction in Changde than FNEA＿KNN.