Research On Methods Of Road Extraction Based On Remote Sensing Image

This paper mainly extracts the roads in remote sensing images.The research process includes: preprocessing the remote sensing image,extracting the features of the road remote sensing image,classifying and identifying the remote sensing image,and extracting the road structure in the road image.Research on remote sensing found that the details and contrast of remote sensing images are not strong enough to extract features.In order to solve this problem,this paper improves the traditional median filtering.By analyzing the peak signal-to-noise ratio,standard deviation and average absolute error of the image before and after improvement,the improved filtering not only smoothes the noise inside the remote sensing image,but also obviously enhanced detail information and contrast of remote sensing images.On the basis of this step,the frequency domain features and texture features of the image are studied.The frequency domain features of the image are extracted by analyzing the spectrum image of the image.The methods based on differential analysis,gray level co-occurrence matrix and Gabor wavelet function are respectively used.Analyze the basic features of the image based on the texture.Then,in order to improve the accuracy of recognition classification,the method of feature fusion is proposed.The weights of each feature are calculated separately,and the single features are weighted and combined to obtain the fusion features.According to the characteristics of the road image extracted in this paper,the support vector machine is used as the final classifier to train the extracted features.The experimental results show that the image classification based on single feature can not accurately identify the road image.In this paper,the accuracy of classification recognition is improved by the algorithm of fusion feature,and the road image identified by classification is further analyzed.According to the characteristics of road remote sensing images,this paper firstly adopts three different methods for three different types of road-based remote sensing images.Firstly,Segmentation of road images using traditional threshold segmentation and genetic algorithm based threshold segmentation.Threshold segmentation based on genetic algorithm not only has a distinctly excellent and time-consuming effect,but also improves the efficiency of segmentation.Then,this paper uses edge detection to segment road images with more obvious edge jumps and improves the Prewitt operator.The latter algorithm is obviously better than other edge algorithms.For more complicated road images,this paper proposes an improved wavelet segmentation algorithm.Experiments show that the algorithm not only has goodrobustness,but also in relatively complex roads.When the remote sensing image is segmented,it shows its good segmentation performance.Based on the characteristics of the initially segmented road binary images,the segmented images are extracted based on the regional features and shape features.Two problems exist in the extraction results of these two methods: the extracted roads have many adhesion areas and the fractures and holes formed by the influence of the surrounding objects,and a fusion feature based on the regional and shape features is proposed.At the same time,the threshold of area area and aspect ratio is increased.A comparative analysis of the three methods of extraction shows that the method of fusion feature extraction can extract the road completely.Then,the extracted road is corrected by the morphological filtering method to smooth the information of the extracted road and fill the fine holes.The experimental results show that the results extracted by the method have high extraction accuracy and strong robustness,and can accurately extract all kinds of roads.