| 摘 要: 文章针对大视场不同尺度的红外目标检测问题，以大视场红外搜索系统为基础，提出了一种基于多级综 合分类器的实时红外目标检测算法，实现由粗到精的虚警剔除。在预处理阶段，该算法首先进行尺度放缩变换，在不 同的尺度上采用Robinson滤波抑制背景，再将不同尺度的滤波结果在原始尺度上融合；在对图像进行背景抑制的基础 上，构造多级串联的综合分类器：在目标的粗检测阶段，采用模糊隶属度融合的分类器剔除大部分虚警；在目标的精细 检测阶段，提取候选目标的特征并进行类间特性分析，设计基于Fisher系数加权的综合分类器以实现真实目标的确认与 虚警剔除。实验表明，该算法能够有效剔除与真实目标特性相似的虚警干扰，对尺度变化的红外运动目标具有较好的检 测效果。 |
| 关键词: Robinson滤波；多级综合分类器；模糊隶属度融合；Fisher判据；红外目标检测 |
|
中图分类号: TP391.4
文献标识码: A
|
|
| Infrared Target Detection Based on the Multi-Level Synthesis Classifier |
|
WANG Peizao,WANG Weihua,WANG Haisong,CHEN Zengping
|
( National University of Defense Technology, Changsha 410073, China)
|
| Abstract: The paper proposes a real-time infrared target detection algorithm based on the multi-level synthetic classifier for the large-field of view infrared search system to achieve false alarm elimination from coarse to fine.In the preprocessing stage,the algorithm firstly transforms the image sequence into different scales and uses the Robinson filtering to suppress the background,then fuses the filter results on the original scale.Based on the background suppression of the image,the paper constructs the multi-level series of the synthesis classifier.In the target coarse detection phase,it uses the fuzzy membership fusion classifier to remove most of the false alarm.In the target fine detection phase,it extracts the characteristics of the candidate targets and carries out the analysis of the inter-class characteristics to design an integrated classifier based on Fisher coefficient weight to realize the real target recognition and false alarm elimination.The experimental results show that the proposed algorithm can effectively eliminate the false alarm with similar characteristics to the real targets.It has a good detection effect on the moving infrared target changing in scale. |
| Keywords: Robinson filtering;multi-level synthesis classifier;fuzzy membership fusion;Fisher criterion;infrared target detection |
