| As the main body of the earth’s ecological system,forests provide a continuous material foundation for human survival and social development.It is not only an important natural resource,but also the largest and most complex ecosystem on the land.It is ensuring forestry production and maintaining ecological balance.As well as dealing with climate warming,etc.,they play an inestimable role.Therefore,timely,accurate and efficient acquisition of forest land change information is extremely important for relevant departments to supervise forest land resources.With the rapid development of remote sensing technology,the emergence of high resolution image provides a new idea for forest land change detection,greatly reducing human and material resources,reducing workload,and improving the efficiency of change detection,which is a milestone for forestland resource management.However,with the advent of the era of big data,massive image data has stricter requirements for change detection accuracy,efficiency,and practicability.Currently,no traditional method is applicable.How to use high-resolution remote sensing images to study an effective method for extracting forestland change information,meeting the requirements of high accuracy of detection results,uniform.detection standards,eliminating human subjective errors,and having an efficient,convenient,and easy-to-operate method for forestland dynamic change detection is particularly urgent.In this study,a local area in Longnan City,Gansu Province,with an area of approximately 84,700 hectares,was used to obtain Sentinel-2A image data of two sceneries in the experimental area.Ground truth sample data for accuracy verification.This article focuses on the above issues and completes the following researches:(1)Combining the current difficulties in the detection of forest land dynamic changes,in view of the influence of the radiation difference between the two phases of remote sensing images,this paper adopts the pseudo-invariant feature method(PIF)in the relative radiation correction,and performs regression analysis on the two phases of images.Correction processing makes the two remote sensing images have the same radiation scale standard.The results show that this method can greatly reduce the radiation difference between the front and rear time phase images due to factors such as the solar altitude angle and acquisition time,and reduce the impact of data quality on the change detection results.(2)This thesis studies forest dynamic change detection methods based on different core algorithms,including pixel-based,feature-based,target-based,and deep learning network model,and elaborates the algorithm principles and analyzes the advantages and disadvantages.At the same time,in view of the shortcomings of the current algorithm,based on the spectral characteristics of high-resolution multispectral image data,a new index NNDWI(New Normalized Difference Water Index)is proposed.The index is mainly in the three bands of red light,near-infrared light and green light.The band calculation is used to increase the separation between vegetation and non-vegetation,and a fixed value is used as a distinguishing parameter through binary processing.Remove the uncontrollability of threshold screening,and unify the classification standard.Taking images of the experimental area as the data source,common vegetation indices such as NDVI,RVI,and NDWI are selected for comparison.The results show that the NNDWI index has the best effect on vegetation extraction,and it is more sensitive to vegetation in the budding period(arable land,early forest land),which can effectively reduce the false change error of cultivated land in forest land change detection.(3)Taking the Sentinel-2A image of a local area in Longnan City,Gansu Province as the data source,elaborated four representative forest land dynamic change detection methods based on the vegetation index method,based on the post-classification comparison method,the ResNet network model,and the NNDWI index method.And compare the results.The experimental results show that this method has the advantages of reducing the spurious changes of cultivated land,quickly dividing the threshold,easy to determine the direction of forest land change,less time,high accuracy,and good applicability.It can provide a certain reference for the detection of forest land dynamic changes. |
