BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第1张图片

论文作者:Ming Kang,Chee-Ming Ting,Fung Fung Ting,Raphaël C. -W. Phan

作者单位:Monash University

论文链接:http://arxiv.org/abs/2309.12585v1

项目链接:https://github.com/mkang315/BGFG-YOLO

内容简介:

1)方向:基于YOLO的目标检测器

2)应用:自动化脑肿瘤检测

3)背景::YOLO(You Only Look Once)基于目标检测器在自动化脑肿瘤检测方面显示出色的准确性。

4)方法:本文提出了一种新颖的BGFG-YOLO架构,将双层路由注意机制(BRA)、广义特征金字塔网络(GFPN)、第四检测头和广义IoU(GIoU)边界框回归损失结合到了YOLOv8中。BGFG-YOLO包含一个注意机制,可以更加关注重要的特征,以及特征金字塔网络,通过将高级语义特征与空间细节合并来丰富特征表示。此外,研究了不同的注意机制、特征融合和检测头架构对脑肿瘤检测准确性的影响。

5)结果:实验结果表明,与YOLOv8x相比,BGFG-YOLO在mAP50上的绝对增加率为3.4%,在脑肿瘤检测数据集Br35H上达到了最先进水平。代码可在https://github.com/mkang315/BGFG-YOLO上获取。

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第2张图片

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第3张图片

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第4张图片

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第5张图片

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第6张图片

BGF-YOLO+脑肿瘤检测:BGF-YOLO: Enhanced YOLOv8 with Multiscale Attentional Feature Fusion for Brain Tumor_第7张图片

你可能感兴趣的:(目标检测,YOLO)