基于卷积神经网络的红外弱小目标检测算法

Infrared Dim Target Detection Based on Convolutional Neural Networks

红外弱小目标成像面积小、可用特征少,目标检测极易受背景杂波干扰,因此,如何准确地检测复杂场景下的弱小目标成为一个技术难点。利用卷积神经网络强大的特征提取能力,基于全卷积神经网络设计单帧红外图像小目标提取网络,准确提取出小目标位置,并基于序列相关性,时空域能量累积增强弱目标的强度,通过自适应滤波器滤除孤立噪点,最终准确地提取出红外弱小目标。实验结果表明,相较于传统算法,本算法信噪比及信噪比增益都最高,比目前实用的Tophat算法提高1以上,结果表明提出的算法提取准确性更高,虚警率更低。

Infrared dim target has small imaging area,less available features and target detection is highly susceptible to background clutter. Therefore,how to detect dim target accurately in complex scenes becomes a technical difficulty. In this paper,based on the powerful feature extraction ability of convolutional neural networks,a single frame infrared small target extraction networks is designed based on fully convolutional neural networks to extract the small target position accurately.Based on the sequence correlation,the energy accumulation in the space-time domain enhances the intensity of the dim target. Filter out the isolated noise through the adaptive filter and finally infrared dim target is extracted accurately.Experimental results show that compared with traditional algorithms,this algorithm has the highest signal-to-noise ratio and signal-to-noise ratio gain,which is higher than 1 than the current practical Tophat algorithm,which proves that the proposed algorithm has higher extraction accuracy and lower false alarm rate.