神经网络 非线性激活层

作用：增强模型的非线性拟合能力

非线性激活层网络：

class activateNet(nn.Module):def __init__(self):super(activateNet,self).__init__()self.relu = nn.ReLU()self.sigmoid = nn.Sigmoid()def forward(self,input):#output = self.relu(input)output = self.sigmoid(input)return outputactivatenet = activateNet()

relu：

sigmoid：

完整代码：

这里是将上一步池化层的输出作为非线性激活层的输入

code：

import torch
import torchvision
from torch import nn
from torch.nn import Conv2d, MaxPool2d
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from torchvision.datasets import ImageFolder
from torchvision import transforms#数据预处理
transform = transforms.Compose([transforms.Resize((224,224)),transforms.ToTensor(),transforms.Normalize(mean = [0.5,0.5,0.5],std = [0.5,0.5,0.5])
])#加载数据集
folder_path = '../images'
dataset = ImageFolder(folder_path,transform=transform)
dataloader = DataLoader(dataset,batch_size=1)#卷积
class convNet(nn.Module):def __init__(self):#调用父类nn.Module的构造函数super(convNet,self).__init__()self.conv1 = Conv2d(in_channels=3,out_channels=6,kernel_size=3,stride=1,padding=0)def forward(self,x):x = self.conv1(x)return xconvnet = convNet()#池化
class poolNet(nn.Module):def __init__(self):super(poolNet,self).__init__()#ceil_mode=True表示边缘不满3x3的部分也会被池化#kernel_size=3 默认是卷积核的大小self.maxpool1 = MaxPool2d(kernel_size=3,ceil_mode=True)self.maxpool2 = MaxPool2d(kernel_size=3,ceil_mode=False)def forward(self,input):output = self.maxpool1(input)#output = self.maxpool2(input)return outputpoolnet = poolNet()#非线性激活层（这里使用relu、sigmoid）
class activateNet(nn.Module):def __init__(self):super(activateNet,self).__init__()self.relu = nn.ReLU()self.sigmoid = nn.Sigmoid()def forward(self,input):#output = self.relu(input)output = self.sigmoid(input)return outputactivatenet = activateNet()writer = SummaryWriter('../logs')cnt = 0
for data in dataloader:img,label = dataprint(img.shape)conv_output = convnet(img)print(conv_output.shape)writer.add_images('input',img,cnt)conv_output = torch.reshape(conv_output,(-1,3,222,222))writer.add_images('conv_output',conv_output,cnt)pool_output = poolnet(conv_output)writer.add_images('pool_output',pool_output,cnt)activate_output = activatenet(pool_output)writer.add_images('activate_output',activate_output,cnt)cnt = cnt + 1writer.close()

神经网络 正则化层

防止过拟合

比较简单 不写了

这这里区别几个概念：

- 归一化：将数据映射到特定区间（如[0,1]或[-1,1]）的过程，常见方法如min-max归一化，通过减去最小值并除以极差实现。其目的是消除不同特征量纲差异的影响，使数据处于统一量级，便于模型快速收敛，例如将图像像素值从0-255转换为0-1。

- 标准化：将数据转换为均值为0、标准差为1的分布，即通过减去均值并除以标准差实现（z-score标准化）。它能让数据更符合正态分布，降低异常值对模型的影响，常用于线性回归、SVM等对数据分布敏感的算法中。

- 正则化：在模型训练中加入额外约束（如L1、L2范数）以防止过拟合的技术。L1正则化通过惩罚权重绝对值，可产生稀疏权重；L2正则化（权重衰减）惩罚权重平方，使权重更平滑。其核心是平衡模型复杂度与拟合能力，提升泛化性能。

神经网络 线性层（全连接层）

输入为上一步非线性激活层的输出

线性层：

code:

class linearNet(nn.Module):def __init__(self):super(linearNet,self).__init__()#in_features(根据下面展开后输出的数值) 表示输入特征的维度，out_features=10 表示输出特征的维度self.linear = nn.Linear(32856,10)def forward(self,input):output = self.linear(input)return outputlinearnet = linearNet()

in_features:

    #为了得到in_featuresprint(torch.reshape(activate_output,(1,1,1,-1)).shape)

完整代码：

code:

import torch
import torchvision
from torch import nn
from torch.nn import Conv2d, MaxPool2d
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from torchvision.datasets import ImageFolder
from torchvision import transforms#数据预处理
transform = transforms.Compose([transforms.Resize((224,224)),transforms.ToTensor(),transforms.Normalize(mean = [0.5,0.5,0.5],std = [0.5,0.5,0.5])
])#加载数据集
folder_path = '../images'
dataset = ImageFolder(folder_path,transform=transform)
#drop_last=True 表示最后一个batch可能小于batch_size，不足的部分会被丢弃
#dataloader = DataLoader(dataset,batch_size=1,drop_last=True)
dataloader = DataLoader(dataset,batch_size=1)#卷积
class convNet(nn.Module):def __init__(self):#调用父类nn.Module的构造函数super(convNet,self).__init__()self.conv1 = Conv2d(in_channels=3,out_channels=6,kernel_size=3,stride=1,padding=0)def forward(self,x):x = self.conv1(x)return xconvnet = convNet()#池化
class poolNet(nn.Module):def __init__(self):super(poolNet,self).__init__()#ceil_mode=True表示边缘不满3x3的部分也会被池化#kernel_size=3 默认是卷积核的大小self.maxpool1 = MaxPool2d(kernel_size=3,ceil_mode=True)self.maxpool2 = MaxPool2d(kernel_size=3,ceil_mode=False)def forward(self,input):output = self.maxpool1(input)#output = self.maxpool2(input)return outputpoolnet = poolNet()#非线性激活层（这里使用relu、sigmoid）
class activateNet(nn.Module):def __init__(self):super(activateNet,self).__init__()self.relu = nn.ReLU()self.sigmoid = nn.Sigmoid()def forward(self,input):#output = self.relu(input)output = self.sigmoid(input)return outputactivatenet = activateNet()#线性层（全连接层）
class linearNet(nn.Module):def __init__(self):super(linearNet,self).__init__()#in_features(根据下面展开后输出的数值) 表示输入特征的维度，out_features=10 表示输出特征的维度self.linear = nn.Linear(32856,10)def forward(self,input):output = self.linear(input)return outputlinearnet = linearNet()writer = SummaryWriter('../logs')cnt = 0
for data in dataloader:img,label = dataprint(img.shape)conv_output = convnet(img)print(conv_output.shape)writer.add_images('input',img,cnt)conv_output = torch.reshape(conv_output,(-1,3,222,222))writer.add_images('conv_output',conv_output,cnt)pool_output = poolnet(conv_output)writer.add_images('pool_output',pool_output,cnt)activate_output = activatenet(pool_output)writer.add_images('activate_output',activate_output,cnt)#为了得到in_featuresprint(torch.reshape(activate_output,(1,1,1,-1)).shape)#flatten:展开成一维数组linear_output = torch.flatten(activate_output)linear_output = linearnet(linear_output)print(linear_output.shape)cnt = cnt + 1writer.close()