Remote Sensing Target Detection And Recognition Based On Hash Representation

The object detection in the remote sensing image is to determine whether or not,one or more objects belonging to the category of interest are included in the given satellite image,based on the data analysis,and to locate the position of each predicted object in the image.With the continuous development of remote sensing technology,object detection of remote sensing images faces more and more challenges,such as large amount of data and high image dimension and so on.Hash learning maps the features of an image from a high dimension to a low dimension and maintains the original similarity.This paper takes remote sensing image as the main research object and conducts in-depth research on object detection of remote sensing image,mainly including the generation of object region proposals,the research of mainstream hash learning algorithm and the targeted improvement of hash learning algorithm for remote sensing image,and apply the improved hash learning algorithm to object detection for remote sensing image.The main innovations of this paper are as follows:(1)In the process of generating object region proposals,based on the preserving similarity and the characteristics of rapid reduction for data of hash learning,we research a weighted voting algorithm for object region proposals based on hash learning.For scenes with complex background and large amount of data,the object region proposals can be effectively sorted to save time and space.(2)In the process of classification based on hash learning,by combining the independence and dependence of data,we raise a method to effectively reduce the amount of data by mapping images to low-dimensional space and ensure high precision at the same time——partial randomness supervised discrete hashing(PRSDH).Due to the characteristics of the hash learning algorithm,most hash learning algorithms compromise between learning efficiency and coding accuracy,which is not suitable for the processing of remote sensing image.Partial randomness supervised discrete hashing effectively avoid such problems,and the coding accuracy is also very high under the premise of ensuring learning efficiency.(3)We conduct in-depth research on the overall process of object detection for remote sensing image,apply the weighted voting algorithm for object region proposals based on hash learning and the partial randomness supervised discrete hashing algorithm to the process of object detection for remote sensing image,and verify the effectivity of this object detection algorithm for remote sensing image through experiments.(4)We apply the proposed object detection algorithm to the practical system,design and implement a complete set of system for object detection and recognition for remote sensing image based on hash learning.The whole system is illustrated in detail from requirements analysis to algorithm implementation,result display and algorithm evaluation.We verify the effectiveness of the proposed object detection algorithm and the algorithm above through experiments again.