受昆虫视觉启发的多光谱遥感影像小目标检测  

Insect visual system inspired small target detection for multi-spectral remotely sensed images

作　　者：黄凤辰[1,2] 李敏[1] 石爱业[1,2] 汤敏[1,2] 徐立中[1,2] 

机构地区：[1]河海大学计算机与信息学院,江苏南京210098 [2]河海大学通信与信息系统工程研究所,江苏南京210098

出　　处：《通信学报》2011年第9期88-95,共8页Journal on Communications

基　　金：国家自然科学基金资助项目(60774092;60901003);高等学校博士学科点专项科研基金资助项目(20070294027)~~

摘　　要：现有多光谱遥感影像目标检测算法大多依赖于结构化背景模型和先验信息,背景复杂化和先验信息匮乏将导致高虚警率的检测结果。受昆虫视觉系统中小目标检测神经元的启发,跳出传统研究思路,提出多光谱遥感影像小目标仿生检测模型及相应的目标检测方法。该方法利用神经元非线性滤波特性对突变信号的敏感性,在局部区域内通过背景纹理抑制和目标边缘增强实现目标检测。实验结果表明该方法在高复杂度背景条件下获得较为稳定的低虚警率检测效果。同时该算法可以较好地平衡背景复杂度和空间分辨率之间的矛盾关系,相比现有检测算法还具有原理简单、易于实现等特点。Most existing target detection algorithms for multi-spectral remotely sensed images dependent on the background model or prior knowledge of spectral,so the false alarm rate of detection algorithms would be enhanced by clutter background and little prior information.Inspired by small target detection neurons of the insect visual system,a bionic small target detection model and its corresponded detection method were proposed for multi-spectral remotely sensed images.Based on the nonlinear filtering characteristic of neural cell which is sensitive to transient signals,target detection was completed by suppressing the local background texture and enhancing target feature.Experimental results showed that the proposed method can detect target with stable low false alarm rate under the condition of complexity background.Meanwhile,the proposed method can balance the contradictory relationship between spatial resolution and background complexity.Compared with existing target detection algorithms,it was simple and easy to be completed.

关 键 词：多光谱遥感影像 目标检测 小目标检测神经元模型 虚警率 

分 类 号：TP71[自动化与计算机技术—检测技术与自动化装置]