Application And Comparison Of Information Extraction And Change Detection Methods In Saline And Alkaline Land

In order to rapidly acquire distributed and widely-distributed data on the dynamic changes of saline-alkali land,this paper firstly combines the 24-characteristic variables such as spectrum,texture,and topography of the image,and uses multi-scale segmentation to use the random forest(RF)classification method to double the time in the study area.Landsat8 OLl images are used for feature extraction and classification processing.Based on this,according to the advantages of high RF accuracy and the availability of important feature information,multiple feature fusion and post-classification comparison methods are used to carry out change detection and method evaluation of saline-alkali land.For image classification,comparative studies were conducted on random forest(RF)and support vector machine classification(SVM).The classification results show that the classification accuracy of RF is slightly higher than that of SVM,the overall classification accuracy is 95.492%,the Kappa coefficient is 0.947,the accuracy of saline-alkali land is 98.510%,and the calculation efficiency is 16.5 times that of SVM,which is more suitable for the target level change of saline-alkali land in the study area.Detection;At the same time,RF results show DEM elevation data,normalized differential vegetation index(NVDI),short wave infrared band,normalized differential humidity index(NDMI)and first principal component mean(ME),etc.that affect the accuracy of the classification model.,Provides important information for follow-up change detection.During the change detection,based on the RF classification results,the change detection was performed using the post-classification comparison method and the multi-feature fusion method,respectively.The combination of post-classification comparison method and ArcGIS cross analysls combined the two-stage classification results by grid transformation,fusion,intersection and other steps to generate land-use transfer matrix and dynamic change maps.The data was quantified to show that the saline-alkaline lands were between 2014 and 2017.The change situation and approximate scope;Based on the important features and importance values provided after RF classification,the multi-feature fusion stage performs average weighted fusion,importance value weighted fusion.and band superposition fusion processing respectively to form a new single feature and vector.Then combined with RF algorithm for change detection.Quantitative analysis and test results show that the efficiency of the two weighted fusion detection methods is approximately twice that of the band-superimposed fusion detection method in the three multi-feature fusion change detection due to the reduction of feature dimensions.The importance value weighting method is slightly higher than the other two methods in terms of detection accuracy,Kappa coefficient,error error,and missing error.The overall detection accuracy is 99.62%,because the change detection is equivalent to the second classification,two levels.The minimum accuracy of the linkage is 94.24%and the highest is 95.13%.Comparison of weighted value-of-importance detection method and post-classification comparison method shows that the two methods are slightly better than the latter in terms of software use complexity,and may be more advantageous in hardware integration,but the quantitative expression is slightly insufficient.Choose according to different needs.In summary,the RF algorithm is combined with the post-classification comparison method and the importance value weighted fusion fusion detection technique.When the saline-alkali soil change detection is completed,the RF classification randomness can be used to satisfy the post-classification comparison method.High-precision dependencies result in quantified and intuitive detection results.At the same time,the feature importance values provided by RF avoid the subjectivity of threshold determination in multi-feature fusion change detection and describe feature attributes more comprehensively.This can be used in practical engineering applications.Reduce integration complexity and improve efficiency.