边缘功能增强:为永久婴儿运动识别产生对抗边缘干扰
Xianfu Bao1, Peng Lin2, Huafei Huang3
1Artificial Intelligence Institute, Hangzhou Dianzi University, Hangzhou, 310018, Zhejiang, China.
概括
一个新的生成边缘指导网络 (GEGN) 改善了新生儿重症监护室 (NICU) 的婴儿姿势估计. 这种方法增强了早产婴儿的四肢运动识别,即使在具有挑战性的低光条件下.
科学领域:
- 医疗成像医学成像
- 计算机视觉 计算机视觉
- 人工智能的人工智能
背景情况:
- 在新生儿重症监护室 (NICU) 估计婴儿姿势对于早产婴儿护理至关重要,但由于照明不佳和数据多样性有限而受到阻碍.
- 现有的模型因早产婴儿的照明和皮肤色调变化而难以提取边缘信息和不规则的像素分布.
研究的目的:
- 开发一种先进的方法来准确地估计婴儿的姿势和肢体运动识别在具有挑战性的NICU环境中.
- 为了解决当前模型在处理低光条件和各种婴儿皮肤色素方面的局限性.
主要方法:
- 提出了生成边缘指导网络 (GEGN),包含边缘重建的自动编码器和对抗边缘扰动分支.
- 集成了一个双损失框架,将交叉和信号相似性损失结合起来,以增强特征表示和学习.
- 引导对抗性扰动以专注于婴儿皮肤区域,以改善特征提取.
主要成果:
- 在Skeleton-V1数据集上获得了95.3%的平均平均精度 (mAP).
- 通过广泛的实验和可视化,在具有挑战性的NICU条件中表现出卓越的性能和稳定性.
- 成功增强低级特征和增加源样本多样性,以改善姿势估计.
结论:
- 在NICU环境中,GEGN方法显著提高了婴儿姿势估计的准确性和稳定性.
- 拟议的方法有效地应对不良照明和早产婴儿不同肤色所带来的挑战.
- 这一进步支持关键医疗干预和为早产婴儿提供实时护理.
相关概念视频
01:30Perceptual Development During Infancy
Perceptual development enables infants to organize and interpret sensory information, helping them understand objects, distances, and surfaces in their environment. Two important aspects are depth perception and haptic perception. These abilities develop as infants gain sensory experience and their nervous systems mature.Development of Depth PerceptionDepth perception is the ability to see objects and surfaces in three dimensions and judge their relative distance. Infants gradually learn to use...
01:29Gestures During Infancy and Toddlerhood
Gestures are nonverbal movements of the hands, arms, or face that allow infants and toddlers to communicate emotions, needs, and intentions before speech develops. Gesture use generally appears before spoken language and decreases as vocabulary expands. For example, an 11-month-old may raise their arms to request being picked up, whereas an 18-month-old may say “up.”Types of GesturesChildren mainly use deictic and representational gestures. Deictic gestures emerge around six months and include...