High-resolution Image Classification And Recognition Method Combining Texture Feature And Super Pixel Segmentation

Remote sensing image classification technology is gradually developing.However,its data source acquisition efficiency,high cost and large data volume make the current remote sensing classification technology unable to meet the needs of current physical classification and feature tracking.The rapid development of drone technology has made it more and more convenient for researchers to obtain high-resolution image data,which makes it difficult to obtain remote sensing image data in the traditional way.Taking the information of eucalyptus as an example,this paper studies the high-resolution and high-resolution images of UAV images,combined with the development of computer in image processing,studies various extraction methods of texture features,and selects gray-scale symbiotic matrix as the extraction texture.The method of eigenvalues is used to extract the correlation features,contrast values,energy values and homogeneity of eucalyptus and non-eucalyptus samples,and to compare the influence factors(image size,step size,direction)affecting the gray level co-occurrence matrix.The test results show that eucalyptus has obvious distinguishable texture characteristics compared with non-eucalyptus.Study multiple color models of color features,and finally select the Lab color model as the image color feature value.Texture feature values and color feature values will be used as the basic data in this paper,and applied in the segmentation method and classification method.By comparing the current image segmentation methods in the computer field(edge detection operator,line extraction method,threshold segmentation,watershed algorithm,region growing and splitting method,super pixel segmentation and DBSCAN clustering method),experiments show that super pixel segmentation combined with DBSCAN aggregation The classification method of the class is more accurate.The method divides the image into different regions according to the color feature value(Lab)and spatial coordinates of the image.These regions have clear boundaries,and the segmentation result is closer to the edge of each class.The resulting image noise is greatly reduced compared to other segmentation methods.In order to further identify eucalyptus,this paper analyzes six supervised classification methods such as parallelepiped.The proposed classification method based on superpixel segmentation and support vector machine is proposed.Based on the segmentation region obtained by superpixel image segmentation,the center of mass point is proposed.Multiple eigenvalues of eucalyptus are extracted and classified as categorical data.Using the above six methods and the method of the present invention,the same test image was used for eucalyptus identification,and the obtained experimental results were quantitatively evaluated.The results show that the classification results obtained by this method are significantly improved compared with the other six classification methods,and the noise generated by the superpixel segmentation method is eliminated.After the experimental comparison,the eucalyptus classification results of this method are more accurate.Finally,the paper proposes a GIS-based implementation of the proposed eucalyptus identification classification method.Firstly,the automatic classification process is studied.The model framework,the classification process flow and the data object relationship are designed for the classification method of this paper.The automatic classification plug-in of Eucalyptus is implemented on the ArcGIS platform through programming method.This paper proposes that the automatic classification and recognition method of segmentation and classification can effectively identify and classify eucalyptus,and provide a new classification idea for the currently indistinguishable features,which provides a good development for the development of high-resolution image classification and recognition of drones.