YOLOv11改进 | 独家创新- 注意力篇 | YOLOv11结合全新多尺度线性注意力机制MLAttention(全网独家创新)

1. MLAttention介绍

(1). 多尺度卷积操作：MLAttention通过多尺度卷积操作来增强不同尺度的特征表达能力。采用了多种卷积核尺寸（例如5x5、1x7、7x1、1x11、11x1、1x21、21x1）的深度可分离卷积来捕捉不同感受野的特征。较小的卷积核擅长捕捉细节信息，而较大的卷积核则能够涵盖更大的上下文信息。这种多尺度的处理方式，确保了网络能够同时对细节和整体信息进行有效的建模。

(2). 多层次的特征融合：MLAttention模块中的多组卷积层会提取不同层次的特征，并通过逐步累加的方式将这些特征进行融合。不同层次的卷积特征通过叠加方式，不仅提高了对复杂特征的捕捉能力，还有效增强了对不同尺度目标的感知能力。这对于复杂场景下的图像特征提取，尤其是包含多尺度目标的场景，有着显著的优势。

(3). 线性注意力机制的引入：MLAttention结合了线性注意力机制，通过生成查询（Q）、键（K）、值（V）三组特征来进行图像局部和全局信息的交互。注意力机制可以通过自适应地学习特征之间的相关性，有效地突出关键区域的特征，同时抑制冗余或不重要的信息。在具体实现中，线性注意力通过Softmax计算注意力权重，然后通过加权求和的方式将重要特征进行增强，从而进一步提高了图像特征提取的准确性和鲁棒性。

(4). 有效的特征增强：MLAttention通过将卷积特征与注意力特征进行融合，在输出时将这些增强后的特征与输入特征进行相加操作。这种设计不仅保留了原始特征的基本信息，还通过注意力机制对特征进行了加强，能够在保留基础信息的同时突出关键特征，进一步提升模型的特征表达能力。

2. 核心代码

import torch
import torch.nn as nn
from torch.nn import functional as Fclass MLAttention(nn.Module):def __init__(self, dim):super().__init__()self.conv0 = nn.Conv2d(dim, dim, 5, padding=2, groups=dim)self.conv0_1 = nn.Conv2d(dim, dim, (1, 7), padding=(0, 3), groups=dim)self.conv0_2 = nn.Conv2d(dim, dim, (7, 1), padding=(3, 0), groups=dim)self.conv1_1 = nn.Conv2d(dim, dim, (1, 11), padding=(0, 5), groups=dim)self.conv1_2 = nn.Conv2d(dim, dim, (11, 1), padding=(5, 0), groups=dim)self.conv2_1 = nn.Conv2d(dim, dim, (1, 21), padding=(0, 10), groups=dim)self.conv2_2 = nn.Conv2d(dim, dim, (21, 1), padding=(10, 0), groups=dim)self.conv3 = nn.Conv2d(dim, dim, 1)# Linear Attentionself.phi_q = nn.Linear(dim, dim)self.phi_k = nn.Linear(dim, dim)self.phi_v = nn.Linear(dim, dim)def forward(self, x):u = x.clone()attn = self.conv0(x)attn_0 = self.conv0_1(attn)attn_0 = self.conv0_2(attn_0)attn_1 = self.conv1_1(attn)attn_1 = self.conv1_2(attn_1)attn_2 = self.conv2_1(attn)attn_2 = self.conv2_2(attn_2)attn = attn + attn_0 + attn_1 + attn_2attn = self.conv3(attn)# Linear AttentionB, C, H, W = x.shapex_flat = x.view(B, C, -1).permute(0, 2, 1)  # (B, N, C)Q = self.phi_q(x_flat)  # (B, N, C)K = self.phi_k(x_flat)  # (B, N, C)V = self.phi_v(x_flat)  # (B, N, C)# 线性注意力：通过Softmax计算权重K_T = K.permute(0, 2, 1)  # (B, C, N)attn_weights = torch.matmul(Q, K_T)  # (B, N, N)attn_weights = F.softmax(attn_weights, dim=-1)attn_output = torch.matmul(attn_weights, V)  # (B, N, C)attn_output = attn_output.permute(0, 2, 1).view(B, C, H, W)  # reshape回原来形状return attn_output * attn + u

3. YOLOv11中添加MLAttention

3.1 在ultralytics/nn下新建Extramodule

3.2 在Extramodule里创建MLAttention

在MLAttention.py文件里添加给出的MLAttention代码

添加完MLAttention代码后，在ultralytics/nn/Extramodule/__init__.py文件中引用

3.3 在tasks.py里引用

在ultralytics/nn/tasks.py文件里引用Extramodule

在tasks.py找到parse_model（ctrl+f 可以直接搜索parse_model位置）

添加如下代码：

        elif m in {MLAttention}:c2 = ch[f]args = [c2, *args]

4. 新建一个yolo11MLAttention.yaml文件

# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLO11 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect# Parameters
nc: 1 # number of classes
scales: # model compound scaling constants, i.e. 'model=yolo11n.yaml' will call yolo11.yaml with scale 'n'# [depth, width, max_channels]n: [0.50, 0.25, 1024] # summary: 319 layers, 2624080 parameters, 2624064 gradients, 6.6 GFLOPss: [0.50, 0.50, 1024] # summary: 319 layers, 9458752 parameters, 9458736 gradients, 21.7 GFLOPsm: [0.50, 1.00, 512] # summary: 409 layers, 20114688 parameters, 20114672 gradients, 68.5 GFLOPsl: [1.00, 1.00, 512] # summary: 631 layers, 25372160 parameters, 25372144 gradients, 87.6 GFLOPsx: [1.00, 1.50, 512] # summary: 631 layers, 56966176 parameters, 56966160 gradients, 196.0 GFLOPs# YOLO11n backbone
backbone:# [from, repeats, module, args]- [-1, 1, Conv, [64, 3, 2]] # 0-P1/2- [-1, 1, Conv, [128, 3, 2]] # 1-P2/4- [-1, 2, C3k2, [256, False, 0.25]]- [-1, 1, Conv, [256, 3, 2]] # 3-P3/8- [-1, 2, C3k2, [512, False, 0.25]]- [-1, 1, Conv, [512, 3, 2]] # 5-P4/16- [-1, 2, C3k2, [512, True]]- [-1, 1, Conv, [1024, 3, 2]] # 7-P5/32- [-1, 2, C3k2, [1024, True]]- [-1, 1, SPPF, [1024, 5]] # 9- [-1, 2, C2PSA, [1024]] # 10# YOLO11n head
head:- [-1, 1, nn.Upsample, [None, 2, "nearest"]]- [[-1, 6], 1, Concat, [1]] # cat backbone P4- [-1, 2, C3k2, [512, False]] # 13- [-1, 1, MLAttention, []]- [-1, 1, nn.Upsample, [None, 2, "nearest"]]- [[-1, 4], 1, Concat, [1]] # cat backbone P3- [-1, 2, C3k2, [256, False]] # 16 (P3/8-small)- [-1, 1, MLAttention, []]- [-1, 1, Conv, [256, 3, 2]]- [[-1, 13], 1, Concat, [1]] # cat head P4- [-1, 2, C3k2, [512, False]] # 19 (P4/16-medium)- [-1, 1, MLAttention, []]- [-1, 1, Conv, [512, 3, 2]]- [[-1, 10], 1, Concat, [1]] # cat head P5- [-1, 2, C3k2, [1024, True]] # 22 (P5/32-large)- [-1, 1, MLAttention, []]- [[17, 21, 26], 1, Detect, [nc]] # Detect(P3, P4, P5)

大家根据自己的数据集实际情况，修改nc大小。

5.模型训练

import warnings
warnings.filterwarnings('ignore')
from ultralytics import YOLOif __name__ == '__main__':model = YOLO(r'D:\yolo\yolov11\ultralytics-main\datasets\yolo11MLAttention.yaml')model.train(data=r'D:\yolo\yolov11\ultralytics-main\datasets\data.yaml',cache=False,imgsz=640,epochs=100,single_cls=False,  # 是否是单类别检测batch=4,close_mosaic=10,workers=0,device='0',optimizer='SGD',amp=True,project='runs/train',name='exp',)

模型结构打印，成功运行：