基于深度学习与多维分形特征提取的海上慢速小目标检测技术

This project addresses the severe challenges caused by robust slow-moving weak target detection in complex sea clutter background. Base on the development trend of maritime radar system as well as the frontier technologies of signal processing theory, this project conducts the research on target detection techniques based on deep learning and multi-dimensional joint fractal properties. Emphasis is placed on sea clutter echo signal modeling in multi-dimensional space, the multi-field fractal and multifractal property analysis of sea clutter, and target detection method based on deep learning and joint fractal properties. By fully exploiting the coupling redundancy among multiple measurements, such as the time, space, frequency and fractal, as well as the advantages multi-dimensional joint processing, this project intends to explore new concepts of deep learning combined with the target detection in multi-dimensional domain and clarifies the principles of information collaborative analysis in the multi-dimensional feature space, and further proposes novel processing criterions and methods. The key scientific issues to be solved include the accurate extraction of feature differences between sea clutter and target in multi-dimensional joint space, and the deep learning network establishment and optimization based on multi-dimensional joint feature. This project seeks to produce innovative research outputs including the multi-field fractal and multifractal property analysis and feature extraction of sea clutter, and target detection method based on multi-dimensional joint features. Therefore, the research results would lay a theoretical foundation for improving the target detection performance of maritime radar system, and provide key technology support for ensuring the target detection capability of radar in complex sea clutter background.

本项目针对复杂海杂波背景对慢速小目标稳健检测造成的严峻挑战，结合对海工作雷达系统的发展趋势，聚焦信号处理理论发展前沿，开展基于深度学习与多维度分形特征的检测技术研究，重点开展基于多重分形的多维度海杂波建模、海杂波多域分形及多重分形特性分析、基于深度学习的联合分形特征检方法研究。通过充分挖掘时间-空间-频率-分形等多维观测域联合处理的优势，探索深度学习与多维特征目标检测相结合的新概念，阐明多维度特征空间中信息协同分析的原理，提出有效的处理新准则、新方法。拟解决的关键科学问题包括多维度联合空间中目标与海杂波的特征差异精确提取问题、基于多维联合特征的深度学习网络建立和优化问题。力求在海杂波多域分形及多重分形特征分析与提取理论、多维度联合特征目标检测技术等方面取得创新性研究成果，为提升对海雷达系统的目标检测性能奠定理论基础，为保障复杂海杂波环境中雷达对海上目标的探测能力提供关键技术支撑

本项目针对复杂海杂波背景对慢速小目标稳健检测造成的严峻挑战，结合对海工作雷达系统的发展趋势，聚焦信号处理理论发展前沿，开展基于深度学习与多维度分形特征的检测技术研究，重点开展基于多重分形的多维度海杂波建模、海杂波多域分形及多重分形特性分析、基于深度学习的联合分形特征检方法研究。通过充分挖掘时间-空间-频率-分形等多维观测域联合处理的优势，探索深度学习与多维特征目标检测相结合的新概念，阐明多维度特征空间中信息协同分析的原理，提出有效的处理新准则、新方法。研究解决了多维度联合空间中目标与海杂波的特征差异精确提取问题、基于多维联合特征的深度学习网络建立和优化问题。本项目在海杂波多域分形及重分形特征分析与提取理论、多维度联合特征目标检测技术等方面取得创新性研究成果，为提升对海雷达系统的目标检测性能奠定理论基础，为保障复杂海杂波环境中雷达对海上目标的探测能力提供关键技术支撑