Yolov5

刺骨的言语ヽ痛彻心扉 2023-01-07 12:29 113阅读 0赞

## 一、改进 ##

与yolov3对比，主要进行了下面4方面改进

（1）输入端：Mosaic数据增强、自适应锚框计算  
（2）Backbone：Focus+CSPDarknet53  
（3）Neck：FPN+PAN  
（4）Prediction：GIOU\_Loss

## 二、模块 ##

### 2.1 自适应锚框 ###

可以选择预先设定的锚框值，也可以在训练过程中，对当前训练集中进行锚框值的计算。

### 2.2 Focus结构 ###

根据[这篇博客][Link 1]解释图像H\*W\*Channel经过focus结构图像变成H‘\*W’\*Channel‘，但是三者乘积不变

如：将32\*32\*3图像经过focus变化后得到16\*16\*12

### 2.3 CSPDarknet53 ###

CSPDarknet53=CSPNet+Darknet53

CSPNet（Cross Stage Partial，跨阶段局部网络），[参考][Link 2]。

基本结构形式：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70][]

Darknet53结构：  
![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 1][]

CSPDarknet53结构：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 2][]

### 2.4 GIOU\_Loss ###

GIOU(Generalized Intersection over Union)

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 3][]

![202101282205446.png][]

**          Ac：两个框的最小闭包区域面积（同时包含了预测框和真实框的最小框的面积）**

**          U：两个框并集区域面积**

![20210128221220660.png][]

## 三、yaml文件分析 ##

yolov5有四种结构：yolov5s、yolov5m、yolov5l、yolov5x，主要对Backbone、Neck中网络的深度及宽度进行修改。

详细参考：

[https://www.cnblogs.com/E-Dreamer-Blogs/p/13297580.html][https_www.cnblogs.com_E-Dreamer-Blogs_p_13297580.html]

[https://blog.csdn.net/mary\_0830/article/details/107124459][https_blog.csdn.net_mary_0830_article_details_107124459]

# #

## 四、yolov5实现安全帽佩戴检测 ##

1、数据集收集并标注

2、搭建yolov5环境，设置参数，模型训练

3、利用pyqt库编写界面

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 4][]

参考：

[深入浅出Yolo系列之Yolov5核心基础知识完整讲解][Link 1]

[Link 1]: https://blog.csdn.net/nan355655600/article/details/107852353?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-2.control&dist_request_id=796b3c54-7649-473a-867a-463441eebb51&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-2.control
[Link 2]: https://blog.csdn.net/haha0825/article/details/106102762
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70]: /images/20221119/11f2a1b8e0264e628f72129e3929ea8a.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20210221195547951.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20210221202405965.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/2021012822432995.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3,size_16,color_FFFFFF,t_70
[202101282205446.png]: https://img-blog.csdnimg.cn/202101282205446.png
[20210128221220660.png]: https://img-blog.csdnimg.cn/20210128221220660.png
[https_www.cnblogs.com_E-Dreamer-Blogs_p_13297580.html]: https://www.cnblogs.com/E-Dreamer-Blogs/p/13297580.html
[https_blog.csdn.net_mary_0830_article_details_107124459]: https://blog.csdn.net/mary_0830/article/details/107124459
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20210221205710674.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3MjY0Mzk3,size_16,color_FFFFFF,t_70