基于YOLOv5+C3CBAM+CBAM注意力的海底生物[海参、海胆、扇贝、海星]检测识别分析系统

news2024/12/23 0:20:53

在我前面的一些文章中也有用到过很多次注意力的集成来提升原生检测模型的性能，这里同样是加入了注意力机制，区别在于，这里同时在两处加入了注意力机制，第一处是讲CBAM集成进入原生的C3模块中，在特征提取部分就可以发挥注意力的作用，另一处是在Detect头前面加入CBAM模块，提升特征融合计算表达的能力，话不多说，首先看下效果图：

改进后的模型yaml文件如下：

#Parameters
nc: 4                           # 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, Conv, [64, 6, 2, 2]],      # 0-P1/2
   [-1, 1, Conv, [128, 3, 2]],        # 1-P2/4
   [-1, 3, C3CBAM, [128]],
   [-1, 1, Conv, [256, 3, 2]],        # 3-P3/8
   [-1, 6, C3CBAM, [256]],
   [-1, 1, Conv, [512, 3, 2]],        # 5-P4/16
   [-1, 9, C3CBAM, [512]],
   [-1, 1, Conv, [1024, 3, 2]],       # 7-P5/32
   [-1, 3, C3CBAM, [1024]],
   [-1, 1, SPPF, [1024, 5]],          # 9
  ]


#Head
head:
  [[-1, 1, Conv, [512, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 6], 1, Concat, [1]],                   # cat backbone P4
   [-1, 3, C3CBAM, [512, False]],               # 13
   [-1, 1, Conv, [256, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 4], 1, Concat, [1]],                   # cat backbone P3
   [-1, 3, C3CBAM, [256, False]],               # 17 (P3/8-small)
   [-1, 1, Conv, [256, 3, 2]],
   [[-1, 14], 1, Concat, [1]],                  # cat head P4
   [-1, 3, C3CBAM, [512, False]],               # 20 (P4/16-medium)
   [-1, 1, Conv, [512, 3, 2]],
   [[-1, 10], 1, Concat, [1]],                  # cat head P5
   [-1, 3, C3CBAM, [1024, False]],              # 23 (P5/32-large)
   [-1, 1, CBAM, [1024]], 

   [[17, 20, 24], 1, Detect, [nc, anchors]],    # Detect(P3, P4, P5)
  ]

这里的改进主要体现在BackBone和Head两部分。

首先是BackBone部分，如下：

其次是Head部分，如下：

这里的改动还是比较多的，不过理清逻辑对照着来处理就行了。

简答看下数据集，如下：

YOLO格式标注数据如下所示：

实例标注内容如下所示：

3 0.754861 0.383951 0.140278 0.180247
2 0.454167 0.150617 0.080556 0.088889
2 0.370139 0.104938 0.068056 0.076543
1 0.7625 0.892593 0.180556 0.214815
1 0.848611 0.785185 0.161111 0.296296
1 0.665278 0.624691 0.144444 0.237037
3 0.578472 0.520988 0.140278 0.182716
3 0.350694 0.497531 0.134722 0.224691
1 0.439583 0.522222 0.101389 0.160494
2 0.416667 0.660494 0.147222 0.22963
0 0.355556 0.897531 0.194444 0.204938
2 0.280556 0.848148 0.127778 0.17037
2 0.284028 0.732099 0.106944 0.17037
2 0.186806 0.781481 0.1125 0.219753
0 0.165278 0.633333 0.113889 0.08642
0 0.194444 0.597531 0.158333 0.128395
2 0.247917 0.490123 0.095833 0.135802
2 0.211111 0.404938 0.127778 0.182716
2 0.215972 0.307407 0.106944 0.106173
2 0.068056 0.191358 0.075 0.081481
2 0.130556 0.237037 0.127778 0.083951
1 0.119444 0.28642 0.088889 0.108642
0 0.029861 0.367901 0.056944 0.148148
0 0.095833 0.407407 0.088889 0.083951
2 0.113194 0.469136 0.123611 0.11358
1 0.053472 0.644444 0.101389 0.330864
1 0.094444 0.782716 0.152778 0.316049

VOC格式标注数据如下：

实例标注内容如下所示：

<annotation>
    <folder>UWA</folder>
    <filename>00b27110-7b71-4c33-aced-a73527a36cd9.jpg</filename>
    <source>
        <database>The UWA Database</database>
        <annotation>UWA</annotation>
        <image>UWA</image>
    </source>
    <owner>
        <name>YSHC</name>
    </owner>    
    <size>
        <width>720</width>
        <height>405</height>
        <depth>3</depth>
    </size>
    <segmented>0</segmented>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>0</xmin>
            <ymin>128</ymin>
            <xmax>71</xmax>
            <ymax>248</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>82</xmin>
            <ymin>154</ymin>
            <xmax>160</xmax>
            <ymax>248</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>145</xmin>
            <ymin>186</ymin>
            <xmax>225</xmax>
            <ymax>284</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>70</xmin>
            <ymin>284</ymin>
            <xmax>196</xmax>
            <ymax>406</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>200</xmin>
            <ymin>290</ymin>
            <xmax>280</xmax>
            <ymax>406</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>271</xmin>
            <ymin>287</ymin>
            <xmax>351</xmax>
            <ymax>368</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>324</xmin>
            <ymin>136</ymin>
            <xmax>401</xmax>
            <ymax>219</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>294</xmin>
            <ymin>73</ymin>
            <xmax>360</xmax>
            <ymax>127</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>194</xmin>
            <ymin>71</ymin>
            <xmax>266</xmax>
            <ymax>140</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>186</xmin>
            <ymin>7</ymin>
            <xmax>236</xmax>
            <ymax>68</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>echinus</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>632</xmin>
            <ymin>84</ymin>
            <xmax>694</xmax>
            <ymax>156</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>scallop</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>533</xmin>
            <ymin>220</ymin>
            <xmax>579</xmax>
            <ymax>270</ymax>
        </bndbox>
    </object>
    
    <object>        
        <name>scallop</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>576</xmin>
            <ymin>124</ymin>
            <xmax>625</xmax>
            <ymax>184</ymax>
        </bndbox>
    </object>
    
</annotation>

默认执行100个epoch的迭代计算，日志输出如下：