论文学习笔记

目录

Elderly Fall Detection Based on Improved YOLOv5s Network

A Deep Learning-Based Object Detection Scheme by Improving YOLOv5 for Sprouted Potatoes Datasets

Optimizing the Hyperparameter Tuning of YOLOv5 for Underwater Detection

Defect Identification of Adhesive Structure Based on DCGAN and YOLOv5

Application of YOLOv5 Based on Attention Mechanism and Receptive Field in Identifying Defects of Thangka Images

ET-YOLOv5s: Toward Deep Identification of Students’ in-Class Behaviors

SETR-YOLOv5n: A Lightweight Low-Light Lane Curvature Detection Method Based on Fractional-Order Fusion Model

---

Elderly Fall Detection Based on Improved YOLOv5s Network

改进点：

使用非对称卷积块(ACB)卷积模块来代替网络中所有的基本卷积

 骨干网络的剩余结构中加入空间注意机制模块SAM

 改进后架构：

A Deep Learning-Based Object Detection Scheme by Improving YOLOv5 for Sprouted Potatoes Datasets

改进点：

在C3模块中用CrossConv代替Conv

SPP改为SPPF

9-Mosaic数据增强方法

提出了一种新的数据处理方法:加载一张原始图片，随机选择8张图片，将它们组合起来，使用超参数translate、scale、scale对其进行处理。

这种方法的优点是丰富了被检测物体的背景；特别是随机缩放增加了微小目标，在一定程度上提高了网络的鲁棒性。另一方面，使用9个镶嵌数据扩充，模型输入将计算9个图像，这隐式地增加了批量大小，并允许模型快速收敛，并降低了对GPU性能的要求。 

 改进后参数列表、架构：

Optimizing the Hyperparameter Tuning of YOLOv5 for Underwater Detection

改进点：

调整了学习参数

Defect Identification of Adhesive Structure Based on DCGAN and YOLOv5

改进点：

引入DCGAN生成数据集

C3模块之后加入CBAM注意力机制模块

 比较损失函数和改进后结果：

Application of YOLOv5 Based on Attention Mechanism and Receptive Field in Identifying Defects of Thangka Images

改进点：

引入scSE

引入CA注意力机制

损失函数CIOU代替GIOU

 改进后的网络架构：

ET-YOLOv5s: Toward Deep Identification of Students’ in-Class Behaviors

改进点：

ESRGAN从高校真实教室环境的原始图像生成高清图像

增加了一个微小物体检测模块

改进后的网络架构：

SETR-YOLOv5n: A Lightweight Low-Light Lane Curvature Detection Method Based on Fractional-Order Fusion Model

改进点：

引入FFM方法来增强具有低平均亮度、模糊细节和高信噪比的图像

提出SETR-C3Block模块

 

改进后的网络架构：