Typical Ground Object Extraction Of GF-1 Image Based On Spectral Analysis

The spectral feature is an important remote sensing radiation and scattering information.Using the radiation and scattering characteristics of the electromagnetic wave,the valuable information is extracted from the data by means of mathematical statistics or physical model inversion.Material emission and reflection radiation characteristics,can further understand the electromagnetic radiation and the interaction of objects.Spectral database refers to the collection of typical features of the measurement spectrum,can cover a variety of typical ground targets spectral and characteristic parameters of the database When the remote sensing data information is not sufficient to support the application requirements,the need to add the spectrum database,as a model to support the background information of the information,And the use of spectral simulation,the model of the supporting data to explain the application of demonstration processes to promote the use of spectral data and the accuracy of identification,the spectral database will be realized to the spectral knowledge base transformation.The characteristics of the spectral information reflect the changes and differences of the feature itself.At present,we can only use the limited remote sensing data to try to understand and understand the complex earth system,from the microscopic spectral differences to extract more accurate The feature information of the feature is very important,so the remote sensing application is more and more dependent on the feature spectrum.In this paper,we quickly and accurately extract the feature information from the image,identify the different substances,reveal the differences,the sensor band,remote sensing image features,features spectral features of the three aspects of in-depth understanding and analysis.In the study area,the canopy spectrum experiments of Robinia pseudoacacia and Hippophae rhamnoides were carried out,and the collateral GF-1 data were obtained.Based on the measurement of the ground parameters and the construction of the spectral library,the multi-temporal GF-1 remote sensing data were used,Combined with image,a priori knowledge and spectral library method,the spectral classification of the experimental area was discussed by spectral mixing analysis using the ROI of the study area.(1)To carry out the spectral measurements of the typical vegetation of the Yanglinggou in the small watershed of the Loess Plateau,to obtain the spectral data of the representative spectral characteristics of the typical vegetation types in the study area,and to convert the ground gray data DN collected from the field into the target Of the reflectivity data.In order to reduce the existence of some errors in the original data and ensure the validity of the spectral data,the preprocessing of the spectral data collected from the field includes: spectral mean processing,spectral data conversion,Savitzky-Golay spectral smoothing denoising,water vapor absorption band Removed.The spectral analysis of five typical vegetation,such as Robinia pseudoacacia and Hippophae rhamnoides,showed that the five typical vegetation indices in the Yanglinggou area of the Loess Plateau and the general trend of the general vegetation were basically consistent.However,there is a large gap between the reflectivity of Robinia pseudoacacia and Hippophae rhamnoides and other vegetation in the range of 740-1360,and it can be used as a feature recognition band.In order to quantitatively describe the differences in vegetation,a first derivative analysis was carried out.The characteristic parameters of each vegetation are calculated in detail.Finally,the accuracy of the characteristic band is checked by the Euclidean distance.The results also show that the selected characteristic band can distinguish the vegetation effectively.Vegetation spectral characteristics analysis,to carry out the analysis of typical vegetation spectral characteristics,study the typical vegetation type spectral characteristics of differences.Effectively established the ground spectral data.(2)GF-1 satellite remote sensing data for remote sensing information sources,remote sensing images for radiation calibration,atmospheric correction and image fusion.Through the statistical analysis of the characteristic data of GF-1 image data,the best index OIF index is selected by band and the best spectral band is selected according to the feature spectrum of the feature.The result shows that the 4,3,2 band is the best band,Reflecting the best band of the information of the sheep ring ditch.By R,G,B false color synthesis repeated comparison,and ultimately agreed that the combination of RGB432 for the best combination of each area.(3)through the field record of the sample and combined with false color remote sensing images on the various features of the show,visual interpretation of interpretation,in the remote sensing image on the establishment of training area,through repeated adjustments to the study area supervision and classification;The results of supervised classification are classified as a layer in decision tree classification.The object classification is carried out by using object-oriented classification,spectral combination texture classification method and multi-classifier aggregation method.Through spectral excavation,the new band variables were constructed,and the spectral excavation separated the locust forest and seabuckthorn forest.(4)Applying the confusion matrix,calculating the total precision of classification and kappa value,the image classification effect is evaluated.