融合全卷积神经网络和视觉显著性的红外小目标检测

为提高复杂背景和噪声干扰下红外小目标检测性能,提出了融合深度神经网络和视觉目标显著性的单阶段红外小目标检测算法.首先设计了基于编码器-解码器架构的轻量级全卷积神经网络对红外图像进行分割,实现背景抑制和目标增强;然后利用红外小目标的显著性特征进一步抑制虚警;最后采用自适应阈值法分离出小目标.网络结构中通过引入多个下采样层降低计算量并增大感受野;通过引入多尺度特征提升背景抑制能力;通过引入注意力机制提升模型训练效果.在真实红外图像上的测试表明,本文算法在检测率、虚警率和运算时间等方面都优于典型红外小目标检测算法,适合进行复杂背景下的红外小目标检测.

Infrared Small Target Detection Based on Fully Convolutional Neural Network and Visual Saliency

To improve the infrared small targets detection performance under complex background and noise interference,a single-stage infrared small target detection algorithm combining fully convolutional neural network and visual saliency is proposed. First,a lightweight fully convolutional neural network based on encoder-decoder architecture is designed to segment infrared images. The network can suppress the background and enhance targets simultaneously. Then,the saliency features of infrared small targets are used to further suppress false alarms. Finally,an adaptive threshold method is used to extract small targets. In the network structure,multiple subsampling layers are introduced to reduce computation load and increase the receptive field;multiscale features are introduced to improve the background suppression ability;attention mechanism is introduced to improve the training result of the model. Experiments on real infrared images show that the proposed algorithm is superior to the typical infrared small target detection algorithm with respect to detection rate,false alarm rate and computation time,and is suitable for infrared small target detection under complex background.