regular commit

README.md

@@ -15,10 +15,10 @@
   对于不同的网络结构,对原始eeg信号采取了预处理,使其拥有不同的shape:<br>
   LSTM:将30s的eeg信号进行FIR带通滤波,获得θ,σ,α,δ,β波,并将它们进行连接后作为输入数据<br>
   resnet_1d:这里使用resnet的一维形式进行实验,(修改nn.Conv2d为nn.Conv1d).<br>
-  DFCNN:将30s的eeg信号进行短时傅里叶变换,并生成频谱图作为输入,并使用resnet网络进行分类.<br>
+  DFCNN:将30s的eeg信号进行短时傅里叶变换,并生成频谱图作为输入,并使用图像分类网络进行分类.<br>
 
 * EEG频谱图<br>
-  这里展示5个睡眠阶段对应的频谱图,它们依次是wake, stage 1, stage 2, stage 3, REM
+  这里展示5个睡眠阶段对应的频谱图,它们依次是wake, stage 1, stage 2, stage 3, REM<br>
   ![image](https://github.com/HypoX64/candock/blob/master/image/spectrum_wake.png)
   ![image](https://github.com/HypoX64/candock/blob/master/image/spectrum_N1.png)
   ![image](https://github.com/HypoX64/candock/blob/master/image/spectrum_N2.png)
@@ -38,14 +38,14 @@
   | resnet18_1d    | 0.8434               | 0.9627                 | 0.0930     |
   | DFCNN+resnet18 | 0.8567               | 0.9663                 | 0.0842     |
   | DFCNN+resnet50 | 0.7916               | 0.9607                 | 0.0983     |
-<br>
+
 * sleep-edfx(only sleep time)<br>
 
   | Network        | Label average recall | Label average accuracy | error rate |
   | :------------- | :------------------- | ---------------------- | ---------- |
   | lstm           | 0.7864               | 0.9166                 | 0.2085     |
   | resnet18_1d    | xxxxxx               | xxxxxx                 | xxxxxx     |
-  | DFCNN+resnet18 | xxxxxx               | xxxxxx                 | xxxxxx     |
+  | DFCNN+resnet18 | 0.7844               | 0.9124                 | 0.219      |
   | DFCNN+resnet50 | xxxxxx               | xxxxxx                 | xxxxxx     |
-<br>
+
 * CinC Challenge 2018<br>
\ No newline at end of file

data.py

@@ -96,7 +96,7 @@ def ToInputShape(data,net_name,norm=True,test_flag = False):
             result = Normalize(result,maxmin = 1000,avg=0,sigma=1000)
         result = result.reshape(batchsize,1,2700)
 
-    elif net_name in ['resnet18','densenet121','densenet201','resnet101','resnet50']:
+    elif net_name in ['dfcnn','resnet18','densenet121','densenet201','resnet101','resnet50']:
         result =[]
         for i in range(0,batchsize):
             spectrum = DSP.signal2spectrum(data[i])

image/confusion_mat

@@ -57,3 +57,12 @@ confusion_mat:
  [   26   699  3159   775   384]
  [    5   380   971  5377   111]
  [   13    34  1490   191 12379]]
+
+DFCNN+resnet18
+avg_recall:0.7844  avg_acc:0.9124  error:0.219
+confusion_mat:
+[[ 3623   243    23     2     4]
+ [ 1905 12703  2452   657    94]
+ [   54   671  3371   399   468]
+ [   29   379  1454  4759   151]
+ [   40    50  1272   124 12881]]

models.py

@@ -12,6 +12,8 @@ def CreatNet(name):
         return CNN()
     elif name == 'resnet18_1d':
         return resnet18_1d()
+    elif name == 'dfcnn':
+        return dfcnn()
     elif name in ['resnet101','resnet50','resnet18']:
         if name =='resnet101':
             net = torchvision.models.resnet101(pretrained=False)
@@ -67,6 +69,72 @@ class LSTM(nn.Module):
         out = self.out(x)
         return out
 
+class dfcnn(nn.Module):
+    def __init__(self):
+        super(dfcnn, self).__init__()
+        self.layer1 = nn.Sequential(       
+            nn.Conv2d(1, 32, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(32),
+            nn.ReLU(inplace = True),
+            nn.Conv2d(32, 32, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(32),
+            nn.ReLU(inplace = True),
+            nn.MaxPool2d(2),   
+        )
+        self.layer2 = nn.Sequential(         
+            nn.Conv2d(32, 64, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(64),
+            nn.ReLU(inplace = True),
+            nn.Conv2d(64, 64, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(64),
+            nn.ReLU(inplace = True),             
+            nn.MaxPool2d(2),                
+        )
+        self.layer3 = nn.Sequential(         
+            nn.Conv2d(64, 128, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(128),
+            nn.ReLU(inplace = True),
+            nn.Conv2d(128, 128, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(128),
+            nn.ReLU(inplace = True),             
+            nn.MaxPool2d(2),                
+        )
+        self.layer4 = nn.Sequential(         
+            nn.Conv2d(128, 256, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(256),
+            nn.ReLU(inplace = True),
+            nn.Conv2d(256, 256, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(256),
+            nn.ReLU(inplace = True),
+            nn.MaxPool2d(2),                        
+        )
+        self.layer5 = nn.Sequential(         
+            nn.Conv2d(256, 512, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(512),
+            nn.ReLU(inplace = True),
+            nn.Conv2d(512, 512, (3,3), 1, 1, bias=False),
+            nn.BatchNorm2d(512),
+            nn.ReLU(inplace = True),
+            nn.MaxPool2d(2),                        
+        )
+
+        self.out = nn.Linear(512*7*7, 5)   # fully connected layer, output 10 classes
+
+    def forward(self, x):
+        x = self.layer1(x)
+
+        x = self.layer2(x)
+
+        x = self.layer3(x)
+
+        x = self.layer4(x)
+
+        x = self.layer5(x)
+        # x = F.avg_pool1d(x, 175)
+        x = x.view(x.size(0), -1)           
+        output = self.out(x)
+        return output
+
 class CNN(nn.Module):
     def __init__(self):
         super(CNN, self).__init__()

train.py

@@ -48,11 +48,11 @@ weight = torch.from_numpy(weight).float()
 if not opt.no_cuda:
     net.cuda()
     weight = weight.cuda()
-    cudnn.benchmark = True
+    # cudnn.benchmark = True
 # print(weight)
 # time.sleep(2000)
 optimizer = torch.optim.Adam(net.parameters(), lr=opt.lr)
-scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=2, gamma=0.95)
+# scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=2, gamma=0.95)
 criterion = nn.CrossEntropyLoss(weight)
 
 
@@ -127,7 +127,7 @@ for epoch in range(opt.epochs):
 
     # torch.cuda.empty_cache() 
     evalnet(net,signals_eval,stages_eval,epoch+1,plot_result,mode = 'all')
-    scheduler.step()
+    # scheduler.step()
 
     if (epoch+1)%opt.network_save_freq == 0:
         torch.save(net.cpu().state_dict(),'./checkpoints/'+opt.model_name+'_epoch'+str(epoch+1)+'.pth')