CNN Based Land Cover Classification Combining Stratified Segmentation And Fusion Of Point Cloud And Very High-spatial Resolution Remote Sensing Image Data

Traditional and Convolutional Neural Network(CNN)-based Geographic Object-based Image Analysis(GE OBIA)land-cover classification methods prosper in Remote Sensing and generate numerous distinguished achievements.However,a bottleneck emerges and hinders further improvements in classification results,due to the insufficiency of information provided by Very High-Spatial Resolution Images(VHSRI).To be specific,the phenomenon of different objects with similar spectrum and the lack of topographic information(heights)are natural drawbacks of VHSRI.Thus,multisource data steps into people's sight and shows a promising future.Firstly,for data fusion,this paper proposes a StdnDSM method which is actually a digital elevation model derived from DTM(Digital Terrain Model)and DSM(Digital Surface Model)to break through the bottleneck by fusing VHSRI and Cloud Points.It smoothes and betters the fusion of Point Cloud and VHSRIs and thus performs well in follow-up classification.The fusion data then were utilized to perform multiresolution segmentation(MRS)and worked as training data for CNN.Moreover,the Grey-Level Co-occurrence Matrix(GLCM)was introduced for a stratified MRS.Secondly,for data processing,the stratified MRS is more efficient than unstratified MRS,and its outcome result is theoretically more rational and explainable than traditional global segmentation.Eventually,classes of segmented polygons were determined by Majority Voting.Compared to pixel-based and traditional object-based classification methods,Majority Voting strategy has stronger robustness and avoids misclassifications caused by minor misclassified centre points.Experimental analysis results suggested that the proposed method is promising for object-based classification.In this paper,a high-resolution remote sensing image is used as experimental data,and a variety of methods are used for comparative tests,which confirms the feasibility and superiority of the proposed new method.According to the analysis of the experimental results,the geographical object-oriented feature classification method based on convolutional neural network has the following main advantages:(1)The method uses the object obtained by the segmentation result as the analysis unit,thereby avoiding the disadvantages of the pixel-based feature classification method.For example,the salt and pepper phenomenon caused by the mixed pixels is greatlyreduced,which greatly improves the calculation efficiency,saves the calculation resources,and improves the classification accuracy.(2)The addition of convolutional neural network can greatly improve the extraction and utilization of high-level information of high spatial resolution remote sensing images,thereby improving the accuracy of ground classification of high-resolution remote sensing images.(3)The multi-scale convolutional neural network can be used to fully extract the multi-scale high-level information in the image,reduce the effect of the scale effect on the classification result,and reduce the difficulty of extracting and using the multi-scale function of the high-resolution image.The accuracy can be solved,and the classification of high-score images can be further improved.(4)The combination of lidar point cloud data and optical remote sensing data can greatly enhance the geometric information of the image,improve the efficiency of multi-scale segmentation,and ultimately improve the effect of ground segmentation.This paper presents a new idea of combining LiDAR and optical remote sensing,and combines convolutional neural network technology with object-oriented remote sensing image classification in a new way,which has a positive and positive significance for future research.