A UAV image segmentation method considering the three-dimensional and edge shape features of buildings

Abstract

A UAV image segmentation method that takes into account the three-dimensional and edge shape features of buildings, including setting segmentation scale parameters according to the maximum object area to be extracted, masking invalid areas, and inputting orthorectified images, elevations A single pixel of an orthophoto image and a SLIC label image is regarded as an object, and the segmentation process is initialized; for the initial pixel object, adjacent pixel objects are found, and if the SLIC label is the same, they are merged until all pixel objects with the same label are merged together. ; Add Canny edge line information and vegetation mask information, and perform edge labeling and vegetation labeling on a single superpixel segmentation object in the superpixel pre-segmentation result; loop iterations, find the most similar object under various constraints, and judge Whether it is appropriate to merge the most similar object with the current object, repeat iterations until there is no object that can continue to be merged; optimize the merge for the small area that has not been merged in the result.

Description

technical field [0001] The invention relates to the technical field of remote sensing image processing, in particular to a method for segmenting images of drones that takes into account the three-dimensional and edge shape features of buildings. Background technique [0002] With the improvement of spatial resolution of remote sensing images, Object Based Image Analysis (OBIA) has gradually become an effective high spatial resolution image analysis tool. Object-level image analysis can weaken the spectral differences inside the object and reduce the salt and pepper noise in the results. Image segmentation is a decisive step in object-level image analysis such as object recognition, information extraction, or image classification. The flying height of UAVs is generally low, and the spatial resolution of the image is very high, which is closer to the close-range image, so the individual ground objects are clearer, the spectral differences inside the ground objects become larg...

AI Technical Summary

Problems solved by technology

The traditional segmentation method based on spectral and shape heterogeneity will encounter greater uncertainty due to the "average" of the features of the segmented object during the merging process. The scale parameters in the segmentation method are difficult to choose and have no definite physical meaning. The edge shape features in the image are difficult to use and the 3D elevation information is only used as another 1D "2D" band information, which may cause the building area and non-building area in the segmentation result to be mistakenly merged, making it difficult for subsequent object-oriented Image analysis generates the correct analysis object

Images