Research On Forest Land Change Detection Based On Object Oriented Classification Of G-F1

Forests are the lungs of the earth.Monitoring,management and construction of forest resources are very important tasks.The traditional artificial field survey method has the disadvantages of time-consuming and laborious effects,and remote sensing technology is characterized by duplication,continuity,high efficiency,and has become the most basic technology in the monitoring of forest resources.With the rise of high-resolution images,object-oriented technology came into being and became a research hotspot in remote sensing technology.This paper aims to study the best method for the change detection after object-oriented classification,preprocessing,segmentation,object-oriented classification,object-oriented classification and change detection of G-Fl in Guangxi State-owned Gaofeng Forest Farm.The results are as follows.(1)Comparing the three fusion methods of PC,GS and NND in the preprocessing of image,comparing the mean,standard deviation,average gradient and information entropy,and combining visual interpretation with objective and subjective evaluation of the three methods respectively.Find out how best the PC fusion method maintains the spectral features,texture features,and geometric features of the image,which helps in image segmentation and classification.(2)Two-level multi-scale segmentation is performed on the image,layer 1 distinguishes between vegetation and non-vegetation,and layer 2 distinguishes tree species within the vegetation.The segmentation scales were selected to be 250,300,and 350;the shape factors were selected to be 0.1 and 0.2 for comparative study;then the visual segmentation method was used to refer to the second survey data to analyze the segmentation results and find the segmentation of layer 1 and layer 2.Scale,shape parameters,and tightness were best at 350,0.1,0.5,and 300,0.2,and 0.8,respectively.(3)When the object-oriented image is classified,the fuzzy algorithm and the classifier algorithm are used to select the nearest neighbor classification method and decision tree classification method respectively.In order to further compare experiments,the cart classifier classification experiment combined with vector data is performed.After the accuracy evaluation,the average of the Kappa coefficient and the overall accuracy of the three are 0.63,0.7211 and 0.715,0.8023 and 0.81,0.8821,respectively.Decision tree classification is superior to the nearest neighbor classification,and combined vector data classification is better than decision tree classification.This classification method can effectively reduce the dislocation between vegetation and provide better imagery for high resolution remote sensing images.Classification direction.(4)Direct detection method and post-classification detection method were performed.For the post-classification detection method,the idea of overlay analysis is used to obtain new objects for the change layer.After the accuracy evaluation,it was found that the direct detection method was obviously at a disadvantage and was not adopted;the overall accuracy and Kappa coefficient of the nearest neighbor classification comparison method,decision tree classification comparison method,and combined vector data classification comparison method were 0.70,4225,0.77,0.5184,respectively.0.83,0.6412.Combined with vector data classification and comparison method,the missed rate and false detection rate are also the lowest.Therefore,in the object-oriented classification detection method,the optimal detection method is combined with vector data classification detection method,which can effectively avoid the errors caused by visual interpretation of the selected sample.