Biologically Inspired Progressive Enhancement Target Detection from Heavy Cluttered SAR Images
Biologically Inspired Progressive Enhancement Target Detection from Heavy Cluttered SAR Images
复制标题
DOI：
10.1007/s12559-016-9405-9
复制
发表时间：
2016-04
影响因子：
5.4
通讯作者：
F. Gao;Fei Ma;Yaotian Zhang;J. Wang;Jinping Sun;Erfu Yang;A. Hussain
中科院分区：
文献类型：
--
作者：
F. Gao;Fei Ma;Yaotian Zhang;J. Wang;Jinping Sun;Erfu Yang;A. Hussain
高分辨率合成孔径雷达（SAR）可以为目标探测提供丰富的信息源，大大增加目标特征的种类和数量。如何有效地从大量SAR图像中提取出感兴趣的目标是目前研究的主要问题。受生物视觉系统的启发，研究者们提出了各种基于生物视觉的目标检测模型，如经典的显著图模型和HMAX模型。但这些方法仅对视觉系统中的视网膜或视皮层进行建模，限制了其对目标特征的提取和整合能力，在复杂环境下检测精度和效率容易受到影响。基于对生物视觉系统中视网膜和视皮层的分析，提出了一种渐进增强的SAR目标检测方法。检测过程分为RET、PVC和AVC三个阶段，分别模拟视网膜、初级和高级视皮层的信息处理链。RET阶段负责消除输入SAR图像的冗余信息，增强输入SAR图像的特征，并将其转换为激励信号。PVC阶段通过神经元之间的竞争机制和特征的结合获得初步特征，进而完成粗检测.在AVC阶段，感受野越大的神经元合成越精确的高级特征，完成最终的精细检测。实验结果表明，该方法在复杂场景下具有较好的检测效果。
High-resolution synthetic aperture radar (SAR) can provide a rich information source for target detection and greatly increase the types and number of target characteristics. How to efficiently extract the target of interest from large amounts of SAR images is the main research issue. Inspired by the biological visual systems, researchers have put forward a variety of biologically inspired visual models for target detection, such as classical saliency map and HMAX. But these methods only model the retina or visual cortex in the visual system, which limit their ability to extract and integrate targets characteristics; thus, their detection accuracy and efficiency can be easily disturbed in complex environment. Based on the analysis of retina and visual cortex in biological visual systems, a progressive enhancement detection method for SAR targets is proposed in this paper. The detection process is divided into RET, PVC, and AVC three stages which simulate the information processing chain of retina, primary and advanced visual cortex, respectively. RET stage is responsible for eliminating the redundant information of input SAR image, enhancing inputs’ features, and transforming them to excitation signals. PVC stage obtainsprimary featuresthrough the competition mechanism between the neurons and the combination of characteristics, and then completes therough detection. In the AVC stage, the neurons with more receptive field compound more preciseadvanced features, completing the finalfine detection. The experimental results obtained in this study show that the proposed approach has better detection results in comparison with the traditional methods in complex scenes.