当前位置: 首页>博士论文>资源详情
光学遥感图像有效区域在轨实时检测与压缩技术研究
中文摘要

随着国防与经济的发展,相关行业应用对航天光学遥感能力提出了更高要求,持续追求遥感图像更高的空间、光谱、时间及辐射分辨率,导致在轨光学遥感图像的数据量呈现指数增长。传统在轨图像处理模式,在卫星有限数传带宽情况下,无法解决卫星每轨成像时间短、在轨图像数据量大、下传图像有效率低与用户需求之间的矛盾,严重制约了航天光学遥感技术发展与应用。本文针对光学遥感卫星发展与应用面临的瓶颈问题,围绕用户应用将遥感图像划分为有效区域(陆地无云区域、海上舰船)和无效区域(陆地云覆盖区域、无舰船海面)两类,面向大幅提升光学遥感卫星在轨使用效能,以提高卫星数传图像数据有效性和增加卫星每轨成像时间为目的,提出了遥感图像有效区域在轨实时检测与压缩方法,并对关键技术进行了攻关,通过算法仿真与硬件实验测试验证了方法的有效性。论文研究是突破已有在轨图像数据处理方法的新尝试,研究成果可为未来光学遥感图像在轨处理提供方法与技术支撑。 首先对光学遥感图像在轨压缩发展现状进行了归纳总结,得出传统的图像压缩方法不能很好的解决遥感成像及应用与有限数传带宽之间的矛盾,提出将在轨实时检测与压缩相结合,检测并剔除遥感图像中的无效区域,提高星地数传图像数据有效性,并增加卫星相机每轨成像时间。对涉及到的实时检测与压缩等关键技术进行了分析,提炼出本文所要研究的问题。 遥感图像有效区域在轨实时检测与压缩的总体研究方案的提出,一方面是基于对遥感图像特性的分析,提出针对不同场景中的云/舰船区域应采取不同的检测策略;另一方面,以成熟的在轨图像压缩算法及当前星地遥感图像压缩/解压缩处理流程为基础,在有效区域检测与压缩一体化框架内发展新的非规则图像的压缩方法,进而形成具有技术原理可行、工程可实施的整体方案。 光学遥感图像中无效的云层覆盖区域数据占用了大量的存储空间和传输带宽,有必要研究高效、快速、准确的遥感图像在轨云检测技术,支撑在轨实时检测与压缩方案。针对遥感图像中云与下垫面的高性能分类问题,提出了基于特征空间线性降维变换的云检测方法以及最小化支持向量数的分类方法。将云检测转化为云与下垫面的分类问题,基于多维特征向量的线性降维有效去除了特征空间参量间的相关冗余性,并可优化低维特征空间的分类性能。同时,提出了基于最小化支持向量数的分类方法,可实现更优的学习推广性能,能够准确区分云与下垫面。该方法通过特征降维策略,在保证云检测性能的同时能够降低计算量,适于在轨实现。实验结果表明,该方法对于云与下垫面具有较高的分类精度。 海洋场景下的有效区域为海面上的舰船,可通过舰船检测剔除大面积海面无效区域,极大提高星地数传图像数据的有效性。针对遥感图像中舰船的高性能检测问题,提出了基于组件几何特性融合判别的舰船检测方法。首先,融合目标几何信息与灰度分布特性,构建组件树以及A-L准则的疑似目标分割算法,可实现复杂云/海面场景下的目标粗检测,有效抑制复杂背景。然后,基于特征距离判别提出舰船目标检测算法,将粗检测结果与模板库中舰船样本的聚类特征比较,进一步实现虚警的剔除。实验结果表明,算法对海面杂波、岛屿、云层等干扰具有较强的抑制能力,可有效支撑实际应用。 经过检测和提取后的地物图像及舰船图像呈现非规则的特点。针对云剔除后形状非规则图像的压缩问题,提出了一种基于局部边缘上下文的填充方法以及形状自适应SA-CCSDS-IDC编码方法。基于图像空间局部相关特性,构建了基于连通区域标记及边缘生长的无效区域局部边缘上下文填充方法,使形状非规则的图像规则化;同时,在小波变换域处理策略下,改进小波变换及比特平面扫描编码过程,构建了基于形状自适应小波变换和非规则比特平面扫描的形状自适应SA-CCSDS-IDC编码方法。实验结果表明,方法可满足非规则图像处理要求,并具备较高的压缩性能。 面向在轨海量图像数据实时处理和空间环境对设备可靠性的要求,提出了采用商用高性能多核DSP和FPGA为核心的处理硬件方案,通过可靠性加固设计满足空间应用要求。同时,研制了一套基于VPX架构的星载遥感图像实时检测压缩处理工程样机及性能测试平台。采用测试图像数据对处理机的功能和性能进行了测试与评估。测试结果表明,处理机在云检测、舰船检测和非规则图像压缩实时处理方面均满足设计指标要求。相对传统在轨图像数据处理,有效区域在轨实时检测与压缩能够极大提高数传图像数据的有效性,有效增加卫星在轨成像时间,从而大幅提升遥感卫星系统应用效能。该成果为未来在轨检测压缩处理应用奠定了坚实的方法与技术基础。 关键词:遥感卫星:遥感图像;云检测;舰船检测;非规则压缩;实时处理

英文摘要

With the development of the national defense and economy, the related applications have put forward higher requirements on space optical remote sensing. The constant pursuit of the enhancement of the spatial,spectral, temporal and radiometric resolution of images has led to an exponential increase in the amount of data in onboard optical remote sensing images. The traditional remote sensing image processing methods suffer from the contradiction between the large amount of remote sensing images and the limited bandwidth of data transmission, which severely restricts the application and development of high-resolution optical remote sensing technology. In order to solve the bottleneck of the development and application of optical remote sensing satellites, this thesis firstly divides the remote sensing images into valid (nocloud-covered ground area, ship) and invalid regions (cloud-covered ground area, sea surface without ships) based on user applications, then proposes the on-board real-time detection and compression methods for the valid regions and explores the technical approaches for enhancing the effectiveness of the satellite image data transmission and increasing the camera imaging time per satellite orbit. The proposed methods can significantly improve the performance of the optical remote sensing satellite in the case of fixed data transmission bandwidth. Also, this thesis verifies the effectiveness of the proposed methods through simulation and hardware implementation. The research in this thesis is a new attempt to break through the existing on-board image data processing methods, and can provide the method and support for on-board processing of the optical remote sensing images in the future. Firstly, in this thesis we summarize the development of optical remote sensing image on-board compression and conclude that the traditional image compression methods cannot solve the contradiction between the remote sensing image applications and limited data transmission bandwidth. Thus we propose a new solution which combines the on-board real-time detection and compression, and removes the invalid regions to increase effectiveness of the satellite image data transmission and the camera imaging time per satellite orbit. We also analyze the related key technologies and extract the problems to be studied in this theirs. Secondly, we develop an overall research scheme of valid region on-board realtime detection and compression for the remote sensing images, which is designed based on two considerations: One is the characteristic analysis of remote sensing images, which results in a conclusion that utilization of different detection strategies for cloud/ship regions in different scenarios. The other is that the proposed compression method should fully integrate the advantage of the existing compression algorithms and the satellite remote sensing image compression/decompression processing flow. Thus, we develop a new irregular image compression method under the integrated framework of the valid region real-time detection and compression, and present the feasible and realizable overall scheme. The invalid data of cloud-covered area in optical remote sensing images takes up a lot of storage space and transmission bandwidth, therefore, it is necessary to study the efficient, fast and accurate on-board cloud detection technology for remote sensing images,to support on-board real-time detection and compression scheme. In this thesis, for the high-performance classification of clouds and underlying surfaces in remote sensing images, a cloud detection method based on feature-space linear reduceddimensional transformation and a classification method that minimizes the number of support vectors are proposed. We convert the cloud detection into the classification problem between cloud and underlying surface. Based on the linear dimensionality reduction in multidimensional feature space, the correlation redundancy between the feature space parameters is effectively eliminate, and the classification performance of low-dimensional eigen space is also optimized. At the same time, we develop a classification method based on minimizing support vector number, which has better learning generalization performance and can distinguish clouds and underlying surfaces accurately. The proposed method can reduce the amount of computation while maintain the performance of cloud detection through feature dimensionality reduction strategy, and is more suitable for on-board implementation. The valid region of the sea scenario is the ship on the sea surface, and the invalid region of the large area can be removed by ship detection, thus the effectiveness of the downlink bandwidth can be greatly improved. In order to solve the problem of ship high-performance detection in remote sensing images, A ship detection method based on component geometric feature fusion discrimination is proposed. Firstly, considering the fusion of target geometry information and gray level distribution, we propose a suspicious target segmentation algorithm based on component tree and A-L criterion, which can realize the coarse target detection in complex cloud/sea scene, and effectively suppress complex background. Then, ship target detection algorithm based on feature distance discrimination is proposed. We compare the coarse detection results with the clustering features of the ship samples in the template library, further eliminate the number of the false alarms. The experimental results show that the algorithm has a strong ability to suppress sea clutter, islands, clouds and other disturbances, and can effectively support practical applications. After valid region detection and extraction, the images of ground targets and ships show irregular characteristics. For the compression problem of irregular images after cloud extraction, a filling method based on local edge context and a shape-adaptive SA-CCSDS-IDC coding method are proposed. Based on the local correlation property in the image, an invalid region filling method based on connected region labeling and edge growth is constructed to regularize the irregular image. At the same time, the wavelet transform and bit-plane coding are improved under the wavelet domain processing strategy. In the process of scanning and coding, a shape-adaptive SA-CC SDS-IDC coding method based on shape-adaptive wavelet transform and irregular bit-plane scan was constructed. Experimental results show that the method can meet the requirements of irregular image processing and have high compression performance. Finally, considering the requirements of on-board real-time processing of massive images and the device reliability in space environment, we propose a hardware solution by using high performance commercial multi-core DSPs and FPGAs with reliability reinforcement design to meet space application requirements. At the same time, we construct a test platform of on-board remote sensing image real-time detection and compression processing based on VPX architecture. The performance of the processor is tested and evaluated through the test platform with some image data sets. The testing results show that the processor meets the design requirements in cloud detection, ship detection and real-time processing of irregular image compression. Compared with the traditional on-board image data processing, the valid region on-board real-time detection and compression can improve the effectiveness of the transmission of image data, increase the on-board imaging time of the camera, thereby greatly enhancing the application efficiency of the remote sensing satellite systems. The results lay a solid technical foundation for future application of on-board real-time detect and compression processing. Keywords: remote sensing satellite, remote sensing image, cloud detection, ship detection, irregular compression, real-time processing

作者相关
主题相关
看过该书的人还在看哪些书