Vegetation Information Extraction From Remote Sensing Image Using Object-oriented MLP Model

With the continuous improvement of the level of science and technology,commercial remote sensing satellites have been rapidly developed in recent years.Significant progress has been made in high-resolution remote sensing image processing technology.How to use remote sensing technology to easily acquire spatial characteristic information in remote sensing images and quantitatively analyze the characteristics of the surface,it becomes one of the main research hotspots in this field.The development of computer science and remote sensing information technology has diversified methods for geographical extraction information from remote sensing images.Remote sensing technology has continuously introduced methods such as machine learning and artificial intelligence,making this process data normalization and method efficiency.The feature extraction of single remote sensing image has the characteristics of macroscopic ? fast and high efficiency.However,for extracting the feature of multiple remotely sensed image features,a single remote sensing image processing method is generally repeated,which ensures the accuracy of the extraction of the geographical features.However,this process is complicated and slow,resulting in the occupation and waste of some resources.This paper has taken vegetation as an example to address the problems in the process of remote sensing image data processing in actual production,such as the remote sensing image data exceeds the computer processing capability and the slow extraction speed of multiple remote sensing images,and it propose that vegetation information extraction from remote sensing image using object-oriented MLP model.Through the OVS-1A satellite remote sensing imagery data in the "ZhuHai-1" series of commercial satellites,the spatial characteristics of remote sensing images were analyzed,and multiple sets of spatial feature sets and reasonable models were selected.The method of vegetation extraction from multiple remote sensing images with the same orbit and different landscapes was analyzed.Rationality and feasibility provide useful reference for extracting land types from remote sensing technology in the third land survey nationwide.The main contents and conclusions of this research are as follows:(1)Using the 1.98 m spatial resolution remote sensing image data as the main datasource,two-dimensional distribution scatter plots were obtained from the spatial location of remote sensing images,and the visible light band information with a quantized value of10 Bit was used to obtain three-dimensional scatter plots.The spatial data sets Band123 and BandAll are used as basic data for the MLP model.By comparing different spatial feature combinations?mean?extreme value?maximum difference degree,and visible light popping difference vegetation index were added to the original visible light band,and the effects of different spatial feature data sets on the extraction of vegetation information from remote sensing images were analyzed.In the object-oriented classification process,after selecting the appropriate segmentation parameters and spectral segmentation optimization,the salt-and-pepper effect appears less than the multi-scale segmentation only.Different segmentation methods result in different segmentation effects of high-resolution remote sensing images.(2)By observing the number of spatial objects classified and classified,the selection of parameters in the modeling process,and the relationship between the two sets of spatial features and the modeling process,the accuracy,loss function value,and misclassification and leakage pixel amount are used.The kappa coefficient obtained from the composition error matrix.The influence of modeling factors and feature data on the accuracy of the model results was analyzed.It was shown that the method of extracting vegetation information from multiple remote sensing images was the main determinant of the accuracy of vegetation information extraction.For machine learning methods,the increase in dimension of feature data is beneficial to improve the accuracy of model prediction.In different spatial data sets,factors such as the visible light and popping difference vegetation index are added,which is also conducive to model prediction of vegetation information improvement.(3)Using multiple remote sensing image data from the same orbit as the data source,Satisfy the assumptions,the geometric and radiation conditions of the remote sensing image data of the same orbit and different landscapes are exactly the same.In the vegetation information extraction process,multi-scale and spectral segmentation methods are used to process remote sensing image data.The processing results are used as training space feature samples,and a feedforward multi-layer perceptron model using reverse transmission training is used.Extract the vegetation characteristics of the study area in remote sensing image data,and apply this model to another remote sensing image with the same orbit and different locations to obtain spatial feature data.Finally,compare with the object-oriented land classification method of artificial single remote sensing image.The results of the indicators meet the actual production requirements and achieve the purpose of the research.