一种基于注意力机制的无监督商标检索方法

针对现有商标特征提取方法无法有效捕捉重点区域的关键信息和图像标注成本过高的问题，提出一种基于注意力机制的无监督商标检索方法。该方法基于实例区分算法，将注意力模块同时运用在神经网络特征映射层的空间维度和通道维度上，通过对各个通道进行权重的分配以及对空间变换参数的学习，增强无监督网络的特征表达能力。该方法在公开的商标数据集上进行了验证，实验表明检索效果优于传统的商标检索方法，甚至优于一些有监督商标检索方法。

An Unsupervised Trademark Retrieval Method Based on Attention Mechanism

In order to solve the deficiency in capturing key information in key areas and the high cost of image annotation in the existing trademark retrieval methods, this paper proposes an unsupervised trademark retrieval method based on attention mechanism. The method applies the attention module to both the spatial dimension and the channel dimension of the feature map layer in the neural network of instance discrimination. Through assigning weights to each channel and learning the spatial transformation parameters, the unsupervised network improve its ability of extracting feature. We further validate the effectiveness of our method on the public trademark datasets and the experiments demonstrate that the proposed method in the paper is better than the traditional trademark retrieval methods, and even surpasses some supervised trademark retrieval methods.