第四次作业：卷积神经网络 part 3

 

1.HybridSN 高光谱分类网络

参考了两位同学的代码：

class_num = 16

class HybridSN(nn.Module):
  def __init__(self):
    super(HybridSN, self).__init__()
    #三维卷积部分：
    #conv1：（1, 30, 25, 25）， 8个 7x3x3 的卷积核 ==>（8, 24, 23, 23）
    #conv2：（8, 24, 23, 23）， 16个 5x3x3 的卷积核 ==>（16, 20, 21, 21）
    #conv3：（16, 20, 21, 21），32个 3x3x3 的卷积核 ==>（32, 18, 19, 19）
    self.conv1 = nn.Conv3d(1,8,(7,3,3))
    self.conv2 = nn.Conv3d(8,16,(5,3,3))
    self.conv3 = nn.Conv3d(16,32,(3,3,3))
    #二维卷积：（576, 19, 19） 64个 3x3 的卷积核，得到 （64, 17, 17）
    #接下来是一个 flatten 操作，变为 18496 维的向量

    self.conv2d = nn.Conv2d(576,64,(3,3))
    self.fc1 =  nn.Linear(18496,256)
    self.fc2 =  nn.Linear(256,128)
    #最后输出为 16 个节点，是最终的分类类别数。
    self.out = nn.Linear(128, class_num)
           
    self.dropout = nn.Dropout(p=0.4)

  def forward(self,x):
    x = F.relu(self.conv1(x))
    x = F.relu(self.conv2(x))
    x = F.relu(self.conv3(x))

    x = x.view(-1,x.shape[1]*x.shape[2],x.shape[3],x.shape[4])
    x = F.relu(self.conv2d(x))
    x = x.view(x.size(0),-1)
    x = self.fc1(x)
    x = self.dropout(x)
    x = self.fc2(x)
    x = self.dropout(x)
    x = self.out(x)
    return x

# 随机输入，测试网络结构是否通
# x = torch.randn(1, 1, 30, 25, 25)
# net = HybridSN()
# y = net(x)
# print(y.shape)

 

 

 

 

 

 

 

 

 2.

代码参考了https://www.cnblogs.com/xiezhijie/p/13472076.html 同学的博客和知乎文章https://zhuanlan.zhihu.com/p/65459972

class_num = 16

class SEBlock(nn.Module):
  def __init__(self,in_channels,r=16):
    super(SEBlock,self).__init__()
    self.globalAvgPool = nn.AdaptiveAvgPool2d((1,1))
    self.fc1 = nn.Linear(in_channels,round(in_channels/r))
    self.fc2 = nn.Linear(round(in_channels/r),in_channels)
  def forward(self,x):
    out = self.globalAvgPool(x)
    out = out.view(out.shape[0],-1)
    out = F.relu(self.fc1(out))
    out = F.sigmoid(self.fc2(out))
    out = out.view(x.shape[0],x.shape[1],1,1)
    out = x * out
    return out

class HybridSN(nn.Module):
  def __init__(self):
    super(HybridSN,self).__init__()
    self.conv3d1 = nn.Conv3d(1,8,kernel_size=(7,3,3),stride=1,padding=0)
    self.bn1 = nn.BatchNorm3d(8)
    self.conv3d2 = nn.Conv3d(8,16,kernel_size=(5,3,3),stride=1,padding=0)
    self.bn2 = nn.BatchNorm3d(16)
    self.conv3d3 = nn.Conv3d(16,32,kernel_size=(3,3,3),stride=1,padding=0)
    self.bn3 = nn.BatchNorm3d(32)
    self.conv2d4 = nn.Conv2d(576,64,kernel_size=(3,3),stride=1,padding=0)
    self.SElayer = SEBlock(64,16)
    self.bn4 = nn.BatchNorm2d(64)
    self.fc1 = nn.Linear(18496,256)
    self.fc2 = nn.Linear(256,128)
    self.fc3 = nn.Linear(128,16)
    self.dropout = nn.Dropout(0.4)

  def forward(self,x):
    out = F.relu(self.bn1(self.conv3d1(x)))
    out = F.relu(self.bn2(self.conv3d2(out)))
    out = F.relu(self.bn3(self.conv3d3(out)))
    out = F.relu(self.bn4(self.conv2d4(out.reshape(out.shape[0],-1,19,19))))
    out = self.SElayer(out)
    out = out.reshape(out.shape[0],-1)
    out = F.relu(self.dropout(self.fc1(out)))
    out = F.relu(self.dropout(self.fc2(out)))
    out = self.fc3(out)
    return out

 

 

 “Excitation操作可以看成学习到了各个channel的权重系数，从而使得模型对各个channel的特征更有辨别能力，这应该也算一种attention机制。”

 

3.《语义分割中的自注意力机制和低秩重重建》：

（1）语义分割

 

 

 

2. Nonlocal神经网络

 

 

 

 

4.《图像语义分割前沿进展》：

挑战：大小各异、形状复杂、环境多变、数量众多