Abstract: Abstract: A method based on Siamese network and transformer is proposed to address the low accuracy problem of infrared dim small target tracking. First, a multi-feature extraction cascade module is constructed to separately extract the deep features of the infrared dim small target template frame and the search frame, and concatenate them with their corresponding HOG features at the dimension level. Second, a multi-head attention mechanism (transformer) is introduced to perform cross-correlation operations on the template feature map and the search feature map, generating a response map. Finally, the target's center position in the image and the regression bounding box are obtained through the response map upsampling network and bounding box prediction network to complete the tracking of the infrared dim small target. Test results on a dataset of 13,655 infrared images show that compared with the KeepTrack tracking method, the success rate is improved by 5.9 percentage points and the precision rate is improved by 1.8 percentage points. Compared with the TransT tracking method, the success rate is improved by 14.2 percentage points and the precision rate is improved by 14.6 percentage points. This proves that the proposed method has higher tracking accuracy for infrared dim small targets in complex backgrounds.

Key words: Target tracking, Infrared dim targets, Siamese networks, transformer, Multi-feature extraction

摘要： 摘 要: 针对红外弱小目标跟踪准确性较低这一问题，提出一种基于孪生网络和transformer的红外弱小目标跟踪方法。首先，构建多特征提取级联模块用于分别提取红外弱小目标模板帧和搜索帧的深度特征，并将二者分别与其对应的HOG特征进行维度层面的串联；其次，引入多头注意力机制（transformer）进行模板特征图和搜索特征图的互相关操作，生成响应图；最后，通过响应图上采样网络和边界框预测网络，获得目标在图像中的中心位置和回归边界框，完成对红外弱小目标的跟踪。在包含13655张红外图像数据集上的测试结果表明：与KeepTrack跟踪方法相比，成功率提高5.9个百分点、精确率提高1.8个百分点；与TransT跟踪方法相比成功率提高14.2个百分点、精确率提高14.6个百分点。证明所提方法对复杂背景下的红外弱小目标跟踪准确性更高。

关键词: 目标跟踪, 红外弱小目标, 孪生网络, transformer, 多特征提取