何小平, 潘晴, 田妮莉. 基于MDFPF-ResNet的红外手掌图像关键点定位[J]. 微电子学与计算机, 2021, 38(10): 9-14. DOI: 10.19304/J.ISSN1000-7180.2021.0135
引用本文: 何小平, 潘晴, 田妮莉. 基于MDFPF-ResNet的红外手掌图像关键点定位[J]. 微电子学与计算机, 2021, 38(10): 9-14. DOI: 10.19304/J.ISSN1000-7180.2021.0135
HE Xiaoping, PAN Qing, TIAN Nili. Location of key points in infrared palm image based on MDFPF-ResNet[J]. Microelectronics & Computer, 2021, 38(10): 9-14. DOI: 10.19304/J.ISSN1000-7180.2021.0135
Citation: HE Xiaoping, PAN Qing, TIAN Nili. Location of key points in infrared palm image based on MDFPF-ResNet[J]. Microelectronics & Computer, 2021, 38(10): 9-14. DOI: 10.19304/J.ISSN1000-7180.2021.0135

基于MDFPF-ResNet的红外手掌图像关键点定位

Location of key points in infrared palm image based on MDFPF-ResNet

  • 摘要: 手诊作为一种能够推理人体器官健康状况的辅助诊断方法,在中医和智能中医的发展中占据着重要的地位.针对现有的红外图像关键点定位方法存在定位不准确的问题,提出一种多尺度空洞卷积特征金字塔融合的残差网络应用于红外手掌图像关键点定位的方法.在每个特征金字塔融合模块中,采用了空洞卷积的金字塔层级与改进型Bottleneck模块的并联结构,在增大感受野的同时,提升了残差网络的泛化能力;并将多个多尺度空洞卷积特征金字塔融合模块级联以逐步获取高层语义特征,在输出端加入全连接层和Dropout层得到关键点定位坐标.通过实验表明,所提方法在红外手掌九宫格图像上具有更好的定位性能,并具有较强的鲁棒性.

     

    Abstract: As an auxiliary diagnostic method that can infer the health status of human organs, hand diagnosis plays an important role in the development of traditional Chinese medicine and intelligent traditional Chinese medicine. In order to solve the problem of inaccurate positioning of key points in infrared image, a residual network based on multi-scale hole convolution feature pyramid fusion is proposed to locate key points in infrared palm image. In each feature pyramid fusion module, the parallel structure of hole convolution pyramid level and improved bottleneck module is adopted, which increases the receptive field and improves the generalization ability of residual network; multiple multi-scale hole convolution feature pyramid fusion modules are cascaded to gradually obtain high-level semantic features, and full connection layer and dropout layer are added at the output end to obtain high-level semantic features Key point positioning coordinates. The experimental results show that the proposed method has better positioning performance and stronger robustness in infrared palm image.

     

/

返回文章
返回