Research On Detection Of Typical Man-made Objects In High Resolution Optical Remote Sensing Images

In recent years,with the rapid development and improvement of the image sensor technology and aerospace industry,the high-resolution remote sensing image has been used more and more widely in the field of military and civilian applications,and gradually become a hot research topic in the field of computer vision and pattern recognition.With the increased number of remote sensing satellite and the shortened period of satellite revisit cycle,the number of high resolution remote sensing image has increased dramatically;the higher resolution of remote sensing image provides users more interested object information,accompanied by more complex background information.How to extract the interested target-especially the man-mede object which is closely related to human activities – from the complex and changeful object environments efficiently and accurately has always been an important and urgent problem.In the traditional approach,the man-mede object is detected by manual interpretation,which has lower detection accuracy and detection efficiency,and as a result,the traditional detection method can't meet the needs of the current applications.Compared with the traditional approach,a reliable automatic detection algorithm of man-made objects has been the main development direction of current remote sensing image processing.Based on the deep analysis of existing man-made target detection algorithms,the advantages and disadvantages of the current main man-made target detection methods are compared and summarized;then,the target characteristics of the typical man-made target of high-resolution optical remote sensing image are studied deeply;Finally,focused on the large-scale high-resolution remote sensing image,this paper designs and realizes a rapid and automatic man-made target detection algorithm.In the meantime,some valuable results have been achieved,and the innovative research work and results are as follows:1.This paper analyzes the characteristics of the high-resolution remote sensing image and its typical man-made targets,and summarizes general process of remote sensing target detection algorithm.A series of key techniques have been studied in detail,such as candidate regions extraction,feature extraction and target confirmation,etc.All the techniques greatly contribute to the theoretical reference.2.In study of single style artificial target detection,this paper focuses on the detection of ship target in large scale high-resolution remote sensing image,and a fast and accurate ship target detection method is proposed.Firstly,the initial saliency map is extracted by the maximum symmetric surround method,which is based on the visual attention mechanism,and updated to get the final saliency map according to the local similarity via a updating mechanism of cellular automata Secondly,by combining with the ship target inherit feature,this paper meliorates the traditional HOG feature,and proposes a new descriptor,named edge-histogram of oriented gradient(E-HOG).The E-HOG is insensitive to the size of the ship target.Finally,the AdaBoost classifier is used to confirm the real ship targets by eliminating the false alarms.As demonstrated in our experiments,the proposed method cannot only detect the ship target in large scale optical remote sensing image quickly,but also has a high detection accuracy of 97.2%;the detection performance of the proposed method outperforms that of the state-of-the-art methods.3.In the study of multi-style artificial target detection,based on the analysis of the difference of geometric shape between man-made objects and natural ones,this paper improves the traditional line extraction algorithm of phase-grouping and meliorates the K-means clustering algorithm,and proposes a new algorithm to detect man-made objects in high-resolution remote sensing optical images.First,an improved phase-grouping method is employed to extracting lines in the image efficiently;Then,the center points of extracted-lines are treated as processing objects and the lines are classified into K classes by K-means Clustering Algorithm;Finally,man-made objects are picked up based on the number of the lines and the geometric elements in every class.The experimental results show that the method proposed in this paper is valid for detection of man-made objects with the detection precision above 90 percent such as the house,vehicle,ship,runways and so on,and the detection speed reaches 10 frames per-second to the image with 512 pixels 512 pixels.4.The multi-style artificial target detection algorithm is studied and transplanted to the hardware platform.First,according to the hardware characteristics of the FPGA,the structure of the multi-style artificial target detection algorithm is reconstructed and optimized;then this paper designs a new hardware architecture to implement the algorithm on FPGA.The FPGA program contains several modules: Image Preprocess module,Region Grow and Line Segment Extraction module,Line Discrimination module,Line Segment Classification module,Target Confirmation module and DDR2 SDRAM Schedule model;Using the Ping-pong operation,the program saves the neighbor frame of the video into different DDR2 SDRAM,and the image in different DDR2 SDRAM is processed by different modules;the CORDIC algorithm and look-up table method are combined to solve the trigonometric function,inverse trigonometric function and square root function,which guarantees the accuracy and real-time performance of the artificial detection algorithm.