基于全局注意力机制的图像检索算法研究

Research on Image Retrieval Algorithms Based on Global Attention Mechanism

针对图像检索中由于图像尺度变化大、目标相似性等影响检索精度的问题,本文提出了一种基于多特征融合的图像检索算法,采用残差网络(ResNet50)提取图像特征,加入全局注意力机制(Global Attention Mechanism),将网络提取的原始特征与GAM注意力机制提取的特征融合,使图像中的关键部分得到更多的关注,实验证明了所提出的算法具有较高的检索准确率和鲁棒性。

This paper proposes an image retrieval algorithm based on multi feature fusion to address the issues of significant changes in image scale and target similarity that affect retrieval accuracy in image retrieval. The algorithm uses a residual network (ResNet50) to extract image features, adds a Global Attention Mechanism, and fuses the original features extracted by the network with the features extracted by the GAM attention mechanism, so that key parts of the image receive more attention, The experiment has proven that the proposed algorithm has high retrieval accuracy and robustness.