基于融合SPD+BIFPN+CBAM改进YOLOv5的奶牛检测识别分析系统开发

news2025/1/10 2:18:27

在我之前的几篇文章中分别应用了不同的tricks来改进yolov5模型如下：

加入针对小目标的SPD-Conv模块

《yolov5s融合SPD-Conv用于提升小目标和低分辨率图像检测性能实践五子棋检测识别》

加入BIFPN特征融合模块

《基于yolov5s+bifpn实践隧道裂缝裂痕检测》

加入注意力机制模块

《YOLOV5融合SE注意力机制和SwinTransformer模块开发实践的中国象棋检测识别分析系统》

感兴趣的话可以自行移步阅读。

本文主要是实际项目需要，针对摄像机拍摄到的奶牛的数据进行检测，考虑到真实情况复杂的问题，这里考虑将一些优化技术加入到原生的yolov5中以期来提升模型的检测性能。首先看下效果：

这里主要加入了：注意力模块CBAM、特征提取模块BIFPN和小目标优化技术SPD-Conv，基准模型是yolov5s，兼顾精度和速度。

直接看下yaml文件，如下：

#Parameters
nc: 1  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors:
  - [10,13, 16,30, 33,23]  # P3/8
  - [30,61, 62,45, 59,119]  # P4/16
  - [116,90, 156,198, 373,326]  # P5/32


#Backbone
backbone:
  # [from, number, module, args]
  [[-1, 1, Focus, [64, 3]],     # 0-P1/2
   [-1, 1, Conv, [128, 3, 1]],  # 1
   [-1,1,SPD,[1]],   # 2 -P2/4
   [-1, 3, C3, [128]],          # 3
   [-1, 1, Conv, [256, 3, 1]],  # 4
   [-1,1,SPD,[1]],   # 5 -P3/8
   [-1, 6, C3, [256]],          # 6
   [-1, 1, Conv, [512, 3, 1]],  # 7-P4/16
   [-1,1,SPD,[1]],   # 8 -P4/16
   [-1, 9, C3, [512]],          # 9
   [-1, 1, Conv, [1024, 3, 1]], # 10-P5/32
   [-1,1,SPD,[1]],   # 11 -P5/32
   [-1, 3, C3, [1024]],         # 12
   [-1, 1, SPPF, [1024, 5]],    # 13
  ]



#Head
head:
  [[-1, 1, Conv, [512, 1, 1]],                     #14
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],     #15
   [[-1, 9], 1, BiFPND, [256, 256]],               #16 
   [-1, 3, C3, [512, False]],                      #17

   [-1, 1, Conv, [256, 1, 1]],                     #18
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],     #19
   [[-1, 6], 1, BiFPND, [128, 128]],               #20 
   [-1, 3, C3, [256, False]],                      #21 

   [-1, 1, Conv, [512, 3, 2]],                     #22
   [[-1, 17, 9], 1, BiFPNT, [256, 256]],          #23
   [-1, 3, C3, [512, False]],                      #24 

   [-1, 1, Conv, [512, 3, 2]],                     #25
   [[-1, 14], 1, BiFPND, [256, 256]],              #26
   [-1, 3, C3, [1024, False]],                     #27 
   [-1, 1, CBAM, [1024]],                            #28

   [[21, 24, 28], 1, Detect, [nc, anchors]],       #29 Detect
  ]

数据集如下：

YOLO格式标注数据如下：

样例标注实例如下所示：

0 0.065234 0.385764 0.107031 0.095139
0 0.06582 0.539931 0.079297 0.165972
0 0.023633 0.762153 0.046484 0.165972
0 0.16875 0.738889 0.111719 0.177778
0 0.183203 0.663542 0.104688 0.200694
0 0.202344 0.571528 0.079687 0.218056
0 0.258008 0.528125 0.063672 0.189583
0 0.30293 0.500694 0.079297 0.261111
0 0.225195 0.313542 0.118359 0.127083
0 0.422461 0.544792 0.126172 0.135417
0 0.442383 0.636458 0.179297 0.095139
0 0.428711 0.715972 0.185547 0.094444
0 0.406836 0.843056 0.134766 0.223611

VOC格式标注数据如下所示：

样例标注实例如下所示：

<annotation>
    <folder>0000</folder>
    <filename>60a4784df6bcca758ee48d5b.jpg</filename>
    <path>60a4784df6bcca758ee48d5b.jpg</path>
    <source>
        <database>Unknown</database>
    </source>
    <size>
        <width>2560</width>
        <height>1440</height>
        <depth>3</depth>
    </size>
    <segmented>0</segmented>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>102</xmin>
            <ymin>111</ymin>
            <xmax>329</xmax>
            <ymax>573</ymax>
        </bndbox>
    </object>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>1</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>1</xmin>
            <ymin>549</ymin>
            <xmax>275</xmax>
            <ymax>819</ymax>
        </bndbox>
    </object>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>19</xmin>
            <ymin>1062</ymin>
            <xmax>416</xmax>
            <ymax>1322</ymax>
        </bndbox>
    </object>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>497</xmin>
            <ymin>1103</ymin>
            <xmax>1165</xmax>
            <ymax>1373</ymax>
        </bndbox>
    </object>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>1132</xmin>
            <ymin>1076</ymin>
            <xmax>1524</xmax>
            <ymax>1325</ymax>
        </bndbox>
    </object>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>1146</xmin>
            <ymin>1306</ymin>
            <xmax>1621</xmax>
            <ymax>1438</ymax>
        </bndbox>
    </object>
    <object>
        <name>cow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>1527</xmin>
            <ymin>1089</ymin>
            <xmax>1970</xmax>
            <ymax>1435</ymax>
        </bndbox>
    </object>
</annotation>

默认设定100次epoch计算，训练完成后结果目录如下所示：