Based On The Feature Elements Of Intelligent Recognition And Extraction Of Remote Sensing Image

The traditional remote classification method is based on pixels,during image feature classificationmain considerationis like pixels statistical characteristicssuch as the grey value, and is mainly dependent on pixels statistical laws to establish remote sensing image classification rules, the traditional image classification classification cannot take full advantage of image information, and the low classification accuracy,while the presence of the classification accuracy is low, slow speed limitations.Due to the high-resolution remote sensing image contains a wealth of spatial information and texture information, traditional classification methods cannot meet the requirements of the classification.Therefore, this paper adopts the comprehensive consideration of the image spectrum, characteristics such as shape, size, texture, topology information of object-oriented remote sensing image classification, it is contains more semantic information of the object as a processing unit, effectively overcome the traditional classification of "salt and pepper phenomenon", and can achieve better classification results.The key technology of the object-oriented image classification is the segmentation of image data.In this paper,image fusion is use 2m panchromatic image with 10 m multi-spectral image of Tianhui satellite, to form a high spatial resolution, color rich, detailed textures expressing fusion image.Improved watershed segmentation algorithm was adopted to realize the intelligent high-resolution remote sensing image processing, which is mainly calculated by direct fusion image color gradient vector space, morphological gradient imageand use mixed opening and closing reconstruction to correct the gradient, the noise in effective in addition to the gradient map and fine-grained structure, and remove all nonsense in the image of fine texture of small area of the connecting piece.In order to solve the gradient image by too many local minima and pseudo there is extreme value point and make the watershed segmentation appear over-segmentation phenomenon. In this paper, an adaptive threshold selection to mark local extended minimum gradient image, in order to limit the minimum number of points, and the use of mandatory minimum operation in the original color gradient image force extend the minimum marking the local extremum point location, make further correction of the gradient image reconstruction, finally do watershed segmentation of the gradient image, and find the segmentation region similarity will be merged into a small region adjacent regions domain, effectively inhibited the V-S watershed over-segmentation, the segmentation precision is improved.In this paper, characteristics analyze and describe of water, roads, and residential areas, which due to the water in the near infrared reflectance and other features of the reflectivity difference is bigger, can be easily to distinguish the water and other surface features, so use the image object in the average near-infraredwavelengths to extract water; Because of the impact of material on the road in the image have a high brightness value, and present a stripon the image, length-width ratio characteristics can differentiate between roads and buildings, so use the ratio of this feature to extract the road;Residents is to use the 135 ° direction of gray level co-occurrence matrix to extract.On the basis of the improved watershed segmentation, to obtain the segmentation image according to the different image characteristics after using the fuzzy theory to extract target feature to water, roads, and residents.The fuzzy theory to extract the features of the concrete steps include:(1) the classification threshold is determined by used multiple times in order to target object is separated from the other;(2) according to the target feature image characteristic value, choosing the appropriate membership functions;(3), each of the video object characteristic value to the membership function, to calculate the object belongs to one kind of membership degree;(4) the blur.According to the fuzzy algorithm to determine the object of a home, if the object is to all classes of the classification of membership degree is less than set threshold, were divided into classification.In order to verify the object-oriented classification method is compared with the traditional method based on pixels, the high resolution image classification has more advantages, in this paper, the object-oriented classification results and traditional precision evaluation of the classification results, the results show that the object-oriented method in the analysis results are more close to the effect of artificial interpretation, the image in the studied area of classification accuracy is much higher than that of traditional classification based on pixels, the accuracy of the extracted features and attributes with the form of real terrain with high consistency, more practical in the extraction of high resolution image information.