Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (19): 214-219.DOI: 10.3778/j.issn.1002-8331.2006-0238

Previous Articles     Next Articles

Salient Detection Model Based on Channel-Spatial Joint Attention Mechanism

陈维婧,周萍,杨海燕,杨青,陈睿   

  1. 1.桂林电子科技大学 电子工程与自动化学院,广西 桂林 541000
    2.桂林电子科技大学 信息与通信学院,广西 桂林 541000
  • Online:2021-10-01 Published:2021-09-29

通道-空间联合注意力机制的显著性检测模型

CHEN Weijing, ZHOU Ping, YANG Haiyan, YANG Qing, CHEN Rui   

  1. CHEN Weijing, ZHOU Ping, YANG Haiyan, YANG Qing, CHEN Rui

Abstract:

Concerning the problems that exist in computer vision, such as saliency areas highlight unevenness and unclear edges, which lead to poor saliency robustness. To solve this problem, a saliency detection model based on channel-spatial joint attention mechanism is proposed. Firstly, it improves the channel attention mechanism and adds the pixel probability values pixel by pixel in the feature map, so as to better obtain the correlation of information between the channels. Then, it integrates the spatial attention mechanism in parallel with the basis of the channel attention mechanism, and the saliency areas with prominent object are received by weighting the spatial information of the feature map. In addition, to obtain a more fine-grained saliency map, the two feature maps output by the channel and spatial attention mechanism are weighted fusion, which fed back to the channel-spatial joint attention mechanism. Sufficient experiments on public datasets DUTS-TE and SOD demonstrate that the proposed method outperforms the others from the value of F-measure and mean absolute error.

Key words: salient detection, channel attention mechanism, spatial attention mechanism

摘要:

针对显著性区域突出不均匀和边缘不清晰导致显著性检测鲁棒性差等问题,提出了一种通道-空间联合注意力机制的显著性检测模型。改进了一种通道注意力机制,将特征图中的像素概率值逐像素相加以更好的获取通道中层间信息的关联性;在通道注意力机制的基础上并行融入了空间注意力机制,对特征图的空间信息进行加权获得目标突出的显著性区域;将通道注意力机制与空间注意力机制输出的两个特征图加权融合反馈至通道-空间联合注意力机制,从而得到细粒度更高的显著图。实验结果表明,该模型在公开的数据集DUTS-TE和SOD上,使用F-measure和平均绝对误差作为评估标准均优于其他同类模型。

关键词: 显著性检测, 通道注意力机制, 空间注意力机制