基于CNN的海空目标检测  

Sea and Air Target′s Detection Based on CNN

作　　者：刘天华[1] 杨绍清[1] 刘松涛[1] 

机构地区：[1]海军大连舰艇学院,辽宁大连116018

出　　处：《现代电子技术》2008年第9期51-53,56,共4页Modern Electronics Technique

基　　金：国家自然基金项目资助(60572160)

摘　　要：针对海空目标运动速度快,机动频繁,要对其既准又快的识别和跟踪,算法和硬件都要求很高的特点,提出了一种新的基于元胞神经网络(CNN)海空目标检测方法。通过大量的仿真实验证明,CNN与传统的方法如各种梯度算子、形态学、小波等相比,其处理结果更加完整细腻,细节更加突出,有利于提取目标的细微特征,特别是对于以云层、海浪为背景的海空光电目标,能更好地进行目标检测。该方法收敛时间快,适合高速并行信号处理,能满足实时处理的要求,因此在军事上具有较大的应用潜力。As the sea and air target＇s main characteristics are their high moving speed and frequent mobility. It requires high precision to the algorithm and hardwires if we want to recognize and track them fast and exactly. This paper brings forward a new method for the sea and air target＇s detection based on Cellular Neural Network （CNN）. Through a lot of emula-tional experiments, it turns out to be that, the image processed by CNN is more integral exquisite, the derail is more prominent, compared with traditional method, such as various grads operator, morphology, wavelet and so on. It is propitious for extracting the exiguous characterestic of the target,especially for those detected by photoelectric sensor in the background of ocean wave and cloud. Its convergence time is short and suit for processing signals with high speed in parallel, and satisfy the requirement for real time processing. So CNN has good potential in military field.

关 键 词：元胞神经网络 目标检测 SOBEL CANNY 形态学 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]