Yolov5 improvement 6: add small target detection layer

front said ： As the current advanced deep learning target detection algorithm YOLOv5, A large number of trick, But when dealing with some complex background problems , It is still prone to the problem of mistakes and omissions . Subsequent articles , Focus on YOLOv5 How to improve is introduced in detail , The purpose is to provide their own meager help and reference for those who need innovation in scientific research or friends who need to achieve better results in engineering projects .

solve the problem ：YOLOv5 One reason for the poor detection effect of small targets is that the size of small target samples is small , and yolov5 The lower sampling multiple of is relatively large , It is difficult to learn the characteristic information of small targets from deep feature maps , Therefore, it is proposed to add a small target detection layer to detect the shallow feature map and the deep feature map after splicing . Add small target detection layer , It can make the network pay more attention to the detection of small targets , Improve the detection effect .

Fang Law ：

The implementation of this method is very simple and effective , It just needs to be modified yolov5 Model files for yaml Small target detection layer can be added , But after adding the detection layer , The problem is that the amount of computation increases , The speed of reasoning detection is reduced . But for small targets , There are really good improvements , modify yaml file , The feature fusion network needs to be modified .

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

# YOLOv5 head
head:
  [[-1, 1, Conv, [512, 1, 1]],#20*20
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
   [-1, 3, C3, [512, False]],  # 13

   [-1, 1, Conv, [512, 1, 1]], #40*40 14
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 4], 1, Concat, [1]],  # cat backbone P3   80*80
   [-1, 3, C3, [512, False]],  # 17 (P3/8-small)  80*80

   [-1, 1, Conv, [256, 1, 1]],
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
   [[-1, 2], 1, Concat, [1]],  # cat backbone P3
   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)21

   [-1, 1, Conv, [256, 3, 2]],
   [[-1, 18], 1, Concat, [1]],  # cat head P4
   [-1, 3, C3, [256, False]],  # 20 (P4/16-medium)24

   [-1, 1, Conv, [256, 3, 2]],  #22   80*80
   [[-1, 14], 1, Concat, [1]], #23 80*80
   [-1, 3, C3, [512, False]], #24 80*80

   [-1, 1, Conv, [512, 3, 2]],
   [[-1, 10], 1, Concat, [1]],  # cat head P5
   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)

   [[21, 24, 27,30], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
  ]

Step four ： take train.py In this article yaml File can , Start training .

junction fruit ： I have done a lot of experiments on multiple data sets , For different data sets, the effect is different , There are also differences in the methods of adding locations to the same dataset , You need to experiment . Most cases are effective and improved .

Let me know ： Improvement of the loss function of content sharing in the next article . Interested friends can pay attention to me , If you have questions, you can leave a message or chat with me in private

PS：, The method of adding detection layer is not only applicable to improvement YOLOv5, You can also improve others YOLO The Internet , such as YOLOv4、v3 etc. .

Last , I hope I can powder each other , Be a friend , Learn and communicate together .