Search research articles

关于 JoVE

概览领导团队博客 JoVE 帮助中心

作者

出版流程编辑委员会范围与政策同行评审常见问题投稿

图书馆员

用户评价订阅访问资源图书馆顾问委员会常见问题

研究

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments 存档

教育

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual 教师资源中心教师网站

使用条款与条件

相关概念视频

Improving Translational Accuracy

Improving Translational Accuracy

您也可能阅读

相关文章

通过共同作者、期刊和引用图与本文相关的文章。

排序

Same author

Early Anomaly Pre-Warning of Buried Pipelines via Dynamic Acceleration Signals: An ICEEMDAN-LSTM Framework.

Sensors (Basel, Switzerland)·2026

Same author

Intracellular acidification by microbiota-derived valeric acid facilitates trans-kingdom ecology limiting Candida parapsilosis colonization.

Cell host & microbe·2026

Same author

Gut Commensals Regulate the Intestinal Kynurenine Pathway.

bioRxiv : the preprint server for biology·2026

Same author

A microbiota-derived bile acid overcomes antibiotic-induced hyporesponsiveness to immune checkpoint therapy by enhancing CD8 <sup>+</sup> T cell antitumor immunity.

bioRxiv : the preprint server for biology·2026

Same author

Polystyrene microplastics disrupt the blood-testis barrier via CEBPB-driven lysosomal autophagy and induce ferroptosis-like injury in human sperm, compromising embryo development.

Journal of hazardous materials·2026

Same author

Single-cell and spatial transcriptomic analysis reveals PAX5's role and regulatory mechanisms in gastric adenocarcinoma lymph node metastasis.

Journal of translational medicine·2026

Search research articles

相关实验视频

Updated: Jul 12, 2025

Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration

Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration

Published on: February 23, 2017

优化基于Dropkey的Grad-CAM:朝着精确的图像特征定位的方向

Yiwei Liu¹, Luping Tang^1,2, Chen Liao³

¹College of Mechanical and Electrical Engineering, Nanjing Forestry University, Nanjing 210037, China.

Sensors (Basel, Switzerland)

|October 28, 2023

概括

此摘要是机器生成的。

梯度加权类激活映射 (Grad-CAM) 在图像识别中与噪声作斗争. 使用Dropkey改进的Grad-CAM增强了深 convolutional神经网络 (CNN) 模型,提高了抗噪声和特征定位精度.

关键词:

类激活映射类的映射.计算机视觉计算机视觉卷积神经网络是一种卷积神经网络.可以解释的解释性.

更多相关视频

Three-dimensional Super Resolution Microscopy of F-actin Filaments by Interferometric PhotoActivated Localization Microscopy iPALM

Three-dimensional Super Resolution Microscopy of F-actin Filaments by Interferometric PhotoActivated Localization Microscopy iPALM

Published on: December 1, 2016

Sample Drift Correction Following 4D Confocal Time-lapse Imaging

Sample Drift Correction Following 4D Confocal Time-lapse Imaging

Published on: April 12, 2014

相关实验视频

Last Updated: Jul 12, 2025

Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration

Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration

Published on: February 23, 2017

Three-dimensional Super Resolution Microscopy of F-actin Filaments by Interferometric PhotoActivated Localization Microscopy iPALM

Three-dimensional Super Resolution Microscopy of F-actin Filaments by Interferometric PhotoActivated Localization Microscopy iPALM

Published on: December 1, 2016

Sample Drift Correction Following 4D Confocal Time-lapse Imaging

Sample Drift Correction Following 4D Confocal Time-lapse Imaging

Published on: April 12, 2014

科学领域:

计算机视觉计算机视觉
深度学习 (Deep Learning) 是一种深度学习.
人工智能的人工智能

背景情况:

梯度加权类激活映射 (Grad-CAM) 是图像识别中特征本地化的一个关键技术,为神经网络决策提供了洞察力.
基于标准Grad-CAM的深卷积神经网络 (CNN) 模型在抵御大规模噪声干扰方面表现出局限性.

研究的目的:

优化深度CNN模型,以提高抗噪能力和精确的特征定位.
评估Dropkey算法的有效性,以提高Grad-CAM对抗噪声的性能.

主要方法:

通过将Dropkey算法与Grad-CAM集成,开发了一个优化的深度CNN模型.
改进的Grad-CAM在梯度计算之前将注意力机制应用于特征图,引入随机性和掩盖注意力得分.
使用dropout作为比较方法来评估Dropkey算法的有效性.

主要成果:

用Dropkey增强的Grad-CAM深度CNN模型显著提高了对大规模噪声干扰的抵抗力.
在噪声偏差为0.6的情况下,Dropkey增强的ResNet50模型实现了0.878的预测信心,优于其他模型.
优化的模型在可视化受扭曲,低对比度和小物体特征影响的特征方面表现出强大的性能.

结论:

滴钥算法有效地提高了Grad-CAM在深度CNN中对噪声的稳定性.
这种优化的方法可以实现准确的特征本地化和可视化,即使在具有挑战性的图像条件下.
增强的Grad-CAM显示了实际计算机视觉应用的巨大潜力,包括自动驾驶,用于验证模型对环境元素的理解.