Study On Extraction Of Characteristic Crops Planting Structure Information In Ningxia Based On GF-2 Images

Crop planting structure information refers to the spatial distribution and result composition of crops within an area or production unit.It is an external manifestation of agricultural production activities for agricultural land,and it is also the ultimate result of rational and efficient use of natural resources and scientific management of fields.Ningxia Autonomous Region has the advantages of land,light energy,and irrigation of Yellow River,which provides an inherent condition for the growth of characteristic crops of Ningxia(e.g.watermelon,Chinese wolfberry,jujube and grape).Quick and accurate access to planting structure information of characteristic crops is not only an important basis for regional crop monitoring,yield estimation and disaster assessment,but also an important evidence for analyzing the spatial pattern changes of characteristic crop and assessing the impact of regional characteristics on agricultural production.Therefore,it is significant to study the information extraction method of Ningxia characteristic crop planting structure.In recent years,with the continuous development of space technology and remote sensing satellites,more and more scholars have applied remote sensing technology to the extraction of crop planting structure information,especially high spatial resolution remote sensing data,which has opened up high-value agriculture for agricultural remote sensing era.However,there are still some difficulties and challenges in the extraction of crop planting structure information based on high spatial resolution.First,the mainstream medium and low resolution remote sensing data cannot meet the needs of precision agriculture,but the attention and application of high resolution data are generally insufficient.Secondly,the traditional remote sensing information extraction models are all based on medium and low resolution remote sensing data,There is no complete information extraction model for the new high resolution image.Thirdly,Based on high resolution remote sensing data,domestic and abroad scholars have relatively few studies on information extraction of similar Ningxia characteristic crops(e.g.watermelon,Chinese wolfberry,jujube and grape),and the selection of classification models and strategies is difficult to meet the demand and development direction of rapid monitoring,accurate acquisition and real-time decision-making.Based on this,in this paper,under the support of GF-2 remote sensing data and field measured spectral data,the separability of spectral features and texture features of GF-2 remote sensing data is fully exploited,and the SAM model,the maximum likelihood model based on pixel level and the object-oriented support vector machine(SVM)model are established,meanwhile the sensitivity and recognition accuracy of different models for four types of characteristic crops were further analyzed and compared.The main research results are as follows:(1)Field spectral data of characteristic crops(e.g.watermelon,Chinese wolfberry,jujube and grape)in Zhongning County of Ningxia were collected using the PSR-1100 field spectrometer.Pre-processing spectral data acquired in the field,and established a field spectral database of characteristic crops.By calculating the "red edge parameters " and "envelope removal values" of the four types of characteristic crops,the quantitative spectral characteristics of the characteristic crops were further described.(2)The data derive from the remote sensing image of GF-2 satellite PMS camera,and performing a series of preprocessing on the remote sensing image.calculateing the standard deviation and correlation coefficient of each band,last selecting NIR-Red-Greed(432)by the OIF.Based on this,the NDVI and texture characteristics of the four crops were calculated,and the differences in the spectral characteristics of the four characteristic crops on the GF-2 remote sensing image were further analyzed,which providing a theoretical basis for the establishment of the information extraction model.(3)Based on the pixel level,a SAM model was established by using GF-2 remote sensing image and field measured spectral data.At the same time,the classification model of spectral single data source and spectral combined texture was established using the maximum likelihood algorithm.The results shown that the texture features can effectively reduce the the classification error,which improves the model classification accuracy by 11.65% and makes the classification result more accurate.(4)Based on the object level,an object-oriented support vector machine(SVM)classification model was established.It was found that the(45,90)segmentation and consolidation scale had the best effect.When using the SVM to classify,the final optimal C is 36.134,and the best ? is 3.305 through the K-fold cross-validation algorithm,heuristic genetic algorithm,particle swarm optimization algorithm.The accuracy verification results shown that the overall classification accuracy of the object-oriented SVM classification model is 88.82%,and the Kappa coefficient is 0.845.(5)The applicability,advantages and disadvantages of the different models in the extraction of characteristic crop planting structure information were compared and analyzed by different accuracy evaluation indicators such as cartographic accuracy and user accuracy.The study found that the object-oriented SVM classification model has the highest recognition accuracy and best effect.The SAM model has the worst classification effect and lowest classification accuracy.At the same time,the jujube can obtain higher classification accuracy with the support of the maximum likelihood classification algorithm.