The Study On Forest Classification And Landscape Pattern Analysis Based On GF-1 Images

Forest resources play an irreplaceable role in the sustainable development of economy, society and environment. Forest type automatic identification based on remote sensing image is an indispensable part of forest resource investigation and monitoring. In this paper, we take the GF-1 satellite image as the data source, according to the absolute radiometric calibration coefficient and the control parameters of orbit and attitude of the satellite in orbit, we do some preprocessing work in the study area like radiometric calibration, atmospheric correction, ortho-rectifying, image fusion, image mosaic and image cropping. Through the analysis of image band features and remote sensing interpretation, the original band from the three aspects of vegetation index, texture features and terrain factors to carry out the static expansion of the original band, to build a classification system. Use decision tree, support vector machine(SVM) and random forests of forest classification and comparison test, determine the support vector machine(SVM) combined with the optimization method of the random forest classifier model, forest classification accuracy is improved, the conclusions are as follows:(1) The effective image preprocessing can correct the effect of illumination, atmosphere and sensor noise, under the premise of keeping the image spectral characteristics, features internal details of the characterization will be more abundant, the ground resolution will be improved.(2) Select the vegetation index, texture information, terrain factor and other characteristic parameters can make the difference between the forest type spectrum is prominent, and it can improve the accuracy of forest type information extraction.(3) Compared with the classification results of decision tree, support vector machine and random forest classification model, the overall accuracy of random forest classification is the highest, the Kappa coefficient is 0.73, and the support vector machine is second. The overall accuracy of decision tree classification is the lowest Kappa coefficient is 0.55. It is found that the support vector machine combined with the random forest classification model can improve the classification accuracy by 1.57% compared with the random forest model, so that the forest classification results in the study area are enhanced.(4) The analysis of forest landscape pattern in the study area shows that the forest landscape diversity index is larger, the landscape fragmentation degree is higher, the spatial distribution is more uniform.