Some polyp detection methods struggle to adequately extract global and long-range semantic information,leading to lower detection accuracy of polyp targets with significant size variations in the presence of environmental noise.In order to address this issue,a polyp detection network based on multi-scale offset awareness is proposed.Firstly,a multi-scale offset-aware attention module is designed to enhance the extraction and fusion capability of image features by applying attention weighting and offset awareness to features at different scales.Secondly,an asymptotic feature fusion module is designed to adaptively spatially weight and fuse feature maps of different scales,thereby capturing richer contextual informa-tion.Extensive experiments demonstrate that this method achieves detection accuracies of 94.8％,94.6％,and 95.8％on three different polyp datasets,outperforming current main-stream object detection methods.
关键词
医学图像/息肉检测/注意力机制/多尺度特征/特征融合
Key words
medical imaging/polyp detection/attention mechanism/multi-scale feature/fea-ture fusion
维普
万方数据