融合注意力和多尺度扩张卷积的图像显著性目标检测方法分析

Analysis of Combining Attention and Multi-scale Expansion Convolution Method for Images Salient Obiect Detection

阐述一种融合通道注意力机制和多尺度扩张卷积的新颖显著性目标检测方法(AMESNet)。首先,为了重点突出图像显著性目标的边缘语义信息,AMESNet构建了高效通道注意力模块,以自适应的选择跨通道互补信息。其次,为了获取边缘结构清晰并且定位准确的显著图,该方法利用多尺度扩张卷积模块增加模型的感受野范围,提取低级语义信息,进行边界锐化。

This paper describes a novel salient object detection method(AMESNet)based on the fusion of channel attention mechanism and multi-scale expansion convolution.Firstly,in order to highlight the edge semantic information of the image saliency object,AMESNet constructs an efficient channel attention module to adaptively select cross-channel complementary information.Secondly,in order to obtain a significant map with clear edge structure and accurate positioning,this method uses multi-scale expansion convolution module to increase the range of receptive fields of the model,extract low-level semantic information,and sharpen boundaries.