NNDL 作业6：基于CNN的XO识别

最新推荐文章于 2023-03-21 21:15:42 发布
最新推荐文章于 2023-03-21 21:15:42 发布
阅读量44 收藏
点赞数 1
文章标签： 1024程序员节
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/staceey/article/details/127501546
版权

 

目录

一、实现卷积-池化-激活

1. Numpy版本：手工实现 卷积-池化-激活

2.Pytorch版本：调用函数实现 卷积-池化-激活

3. 可视化：了解数字与图像之间的关系

二、 基于CNN的XO识别

1.数据集

2.构建模型

3.训练模型

4. 测试训练好的模型

5.计算模型的准确率

6.查看训练好的模型的特征图

7.查看训练好的模型的卷积核

8.训练模型源代码

9.测试模型源代码

总结

参考

一、实现卷积-池化-激活

1. Numpy版本：手工实现 卷积-池化-激活

import numpy as np
 
x = np.array([[-1, -1, -1, -1, -1, -1, -1, -1, -1],
              [-1, 1, -1, -1, -1, -1, -1, 1, -1],
              [-1, -1, 1, -1, -1, -1, 1, -1, -1],
              [-1, -1, -1, 1, -1, 1, -1, -1, -1],
              [-1, -1, -1, -1, 1, -1, -1, -1, -1],
              [-1, -1, -1, 1, -1, 1, -1, -1, -1],
              [-1, -1, 1, -1, -1, -1, 1, -1, -1],
              [-1, 1, -1, -1, -1, -1, -1, 1, -1],
              [-1, -1, -1, -1, -1, -1, -1, -1, -1]])
print("x=\n", x)
# 初始化 三个 卷积核
Kernel = [[0 for i in range(0, 3)] for j in range(0, 3)]
Kernel[0] = np.array([[1, -1, -1],
                      [-1, 1, -1],
                      [-1, -1, 1]])
Kernel[1] = np.array([[1, -1, 1],
                      [-1, 1, -1],
                      [1, -1, 1]])
Kernel[2] = np.array([[-1, -1, 1],
                      [-1, 1, -1],
                      [1, -1, -1]])
 
# --------------- 卷积  ---------------
stride = 1  # 步长
feature_map_h = 7  # 特征图的高
feature_map_w = 7  # 特征图的宽
feature_map = [0 for i in range(0, 3)]  # 初始化3个特征图
for i in range(0, 3):
    feature_map[i] = np.zeros((feature_map_h, feature_map_w))  # 初始化特征图
for h in range(feature_map_h):  # 向下滑动，得到卷积后的固定行
    for w in range(feature_map_w):  # 向右滑动，得到卷积后的固定行的列
        v_start = h * stride  # 滑动窗口的起始行（高）
        v_end = v_start + 3  # 滑动窗口的结束行（高）
        h_start = w * stride  # 滑动窗口的起始列（宽）
        h_end = h_start + 3  # 滑动窗口的结束列（宽）
        window = x[v_start:v_end, h_start:h_end]  # 从图切出一个滑动窗口
        for i in range(0, 3):
            feature_map[i][h, w] = np.divide(np.sum(np.multiply(window, Kernel[i][:, :])), 9)
print("feature_map:\n", np.around(feature_map, decimals=2))
 
# --------------- 池化  ---------------
pooling_stride = 2  # 步长
pooling_h = 4  # 特征图的高
pooling_w = 4  # 特征图的宽
feature_map_pad_0 = [[0 for i in range(0, 8)] for j in range(0, 8)]
for i in range(0, 3):  # 特征图 补 0 ，行 列 都要加 1 (因为上一层是奇数，池化窗口用的偶数)
    feature_map_pad_0[i] = np.pad(feature_map[i], ((0, 1), (0, 1)), 'constant', constant_values=(0, 0))
# print("feature_map_pad_0 0:\n", np.around(feature_map_pad_0[0], decimals=2))
 
pooling = [0 for i in range(0, 3)]
for i in range(0, 3):
    pooling[i] = np.zeros((pooling_h, pooling_w))  # 初始化特征图
for h in range(pooling_h):  # 向下滑动，得到卷积后的固定行
    for w in range(pooling_w):  # 向右滑动，得到卷积后的固定行的列
        v_start = h * pooling_stride  # 滑动窗口的起始行（高）
        v_end = v_start + 2  # 滑动窗口的结束行（高）
        h_start = w * pooling_stride  # 滑动窗口的起始列（宽）
        h_end = h_start + 2  # 滑动窗口的结束列（宽）
        for i in range(0, 3):
            pooling[i][h, w] = np.max(feature_map_pad_0[i][v_start:v_end, h_start:h_end])
print("pooling:\n", np.around(pooling[0], decimals=2))
print("pooling:\n", np.around(pooling[1], decimals=2))
print("pooling:\n", np.around(pooling[2], decimals=2))
 
 
# --------------- 激活  ---------------
def relu(x):
    return (abs(x) + x) / 2
 
 
relu_map_h = 7  # 特征图的高
relu_map_w = 7  # 特征图的宽
relu_map = [0 for i in range(0, 3)]  # 初始化3个特征图
for i in range(0, 3):
    relu_map[i] = np.zeros((relu_map_h, relu_map_w))  # 初始化特征图
 
for i in range(0, 3):
    relu_map[i] = relu(feature_map[i])
 
print("relu map :\n",np.around(relu_map[0], decimals=2))
print("relu map :\n",np.around(relu_map[1], decimals=2))
print("relu map :\n",np.around(relu_map[2], decimals=2))

运行结果：

x=
 [[-1 -1 -1 -1 -1 -1 -1 -1 -1]
 [-1  1 -1 -1 -1 -1 -1  1 -1]
 [-1 -1  1 -1 -1 -1  1 -1 -1]
 [-1 -1 -1  1 -1  1 -1 -1 -1]
 [-1 -1 -1 -1  1 -1 -1 -1 -1]
 [-1 -1 -1  1 -1  1 -1 -1 -1]
 [-1 -1  1 -1 -1 -1  1 -1 -1]
 [-1  1 -1 -1 -1 -1 -1  1 -1]
 [-1 -1 -1 -1 -1 -1 -1 -1 -1]]
feature_map:
 [[[ 0.78 -0.11  0.11  0.33  0.56 -0.11  0.33]
  [-0.11  1.   -0.11  0.33 -0.11  0.11 -0.11]
  [ 0.11 -0.11  1.   -0.33  0.11 -0.11  0.56]
  [ 0.33  0.33 -0.33  0.56 -0.33  0.33  0.33]
  [ 0.56 -0.11  0.11 -0.33  1.   -0.11  0.11]
  [-0.11  0.11 -0.11  0.33 -0.11  1.   -0.11]
  [ 0.33 -0.11  0.56  0.33  0.11 -0.11  0.78]]

 [[ 0.33 -0.56  0.11 -0.11  0.11 -0.56  0.33]
  [-0.56  0.56 -0.56  0.33 -0.56  0.56 -0.56]
  [ 0.11 -0.56  0.56 -0.78  0.56 -0.56  0.11]
  [-0.11  0.33 -0.78  1.   -0.78  0.33 -0.11]
  [ 0.11 -0.56  0.56 -0.78  0.56 -0.56  0.11]
  [-0.56  0.56 -0.56  0.33 -0.56  0.56 -0.56]
  [ 0.33 -0.56  0.11 -0.11  0.11 -0.56  0.33]]

 [[ 0.33 -0.11  0.56  0.33  0.11 -0.11  0.78]
  [-0.11  0.11 -0.11  0.33 -0.11  1.   -0.11]
  [ 0.56 -0.11  0.11 -0.33  1.   -0.11  0.11]
  [ 0.33  0.33 -0.33  0.56 -0.33  0.33  0.33]
  [ 0.11 -0.11  1.   -0.33  0.11 -0.11  0.56]
  [-0.11  1.   -0.11  0.33 -0.11  0.11 -0.11]
  [ 0.78 -0.11  0.11  0.33  0.56 -0.11  0.33]]]
pooling:
 [[1.   0.33 0.56 0.33]
 [0.33 1.   0.33 0.56]
 [0.56 0.33 1.   0.11]
 [0.33 0.56 0.11 0.78]]
pooling:
 [[0.56 0.33 0.56 0.33]
 [0.33 1.   0.56 0.11]
 [0.56 0.56 0.56 0.11]
 [0.33 0.11 0.11 0.33]]
pooling:
 [[0.33 0.56 1.   0.78]
 [0.56 0.56 1.   0.33]
 [1.   1.   0.11 0.56]
 [0.78 0.33 0.56 0.33]]
relu map :
 [[0.78 0.   0.11 0.33 0.56 0.   0.33]
 [0.   1.   0.   0.33 0.   0.11 0.  ]
 [0.11 0.   1.   0.   0.11 0.   0.56]
 [0.33 0.33 0.   0.56 0.   0.33 0.33]
 [0.56 0.   0.11 0.   1.   0.   0.11]
 [0.   0.11 0.   0.33 0.   1.   0.  ]
 [0.33 0.   0.56 0.33 0.11 0.   0.78]]
relu map :
 [[0.33 0.   0.11 0.   0.11 0.   0.33]
 [0.   0.56 0.   0.33 0.   0.56 0.  ]
 [0.11 0.   0.56 0.   0.56 0.   0.11]
 [0.   0.33 0.   1.   0.   0.33 0.  ]
 [0.11 0.   0.56 0.   0.56 0.   0.11]
 [0.   0.56 0.   0.33 0.   0.56 0.  ]
 [0.33 0.   0.11 0.   0.11 0.   0.33]]
relu map :
 [[0.33 0.   0.56 0.33 0.11 0.   0.78]
 [0.   0.11 0.   0.33 0.   1.   0.  ]
 [0.56 0.   0.11 0.   1.   0.   0.11]
 [0.33 0.33 0.   0.56 0.   0.33 0.33]
 [0.11 0.   1.   0.   0.11 0.   0.56]
 [0.   1.   0.   0.33 0.   0.11 0.  ]
 [0.78 0.   0.11 0.33 0.56 0.   0.33]]

2.Pytorch版本：调用函数实现 卷积-池化-激活

import numpy as np
import torch
import torch.nn as nn
 
x = torch.tensor([[[[-1, -1, -1, -1, -1, -1, -1, -1, -1],
                    [-1, 1, -1, -1, -1, -1, -1, 1, -1],
                    [-1, -1, 1, -1, -1, -1, 1, -1, -1],
                    [-1, -1, -1, 1, -1, 1, -1, -1, -1],
                    [-1, -1, -1, -1, 1, -1, -1, -1, -1],
                    [-1, -1, -1, 1, -1, 1, -1, -1, -1],
                    [-1, -1, 1, -1, -1, -1, 1, -1, -1],
                    [-1, 1, -1, -1, -1, -1, -1, 1, -1],
                    [-1, -1, -1, -1, -1, -1, -1, -1, -1]]]], dtype=torch.float)
print(x.shape)
print(x)
 
print("--------------- 卷积  ---------------")
conv1 = nn.Conv2d(1, 1, (3, 3), 1)  # in_channel , out_channel , kennel_size , stride
conv1.weight.data = torch.Tensor([[[[1, -1, -1],
                                    [-1, 1, -1],
                                    [-1, -1, 1]]
                                   ]])
conv2 = nn.Conv2d(1, 1, (3, 3), 1)  # in_channel , out_channel , kennel_size , stride
conv2.weight.data = torch.Tensor([[[[1, -1, 1],
                                    [-1, 1, -1],
                                    [1, -1, 1]]
                                   ]])
conv3 = nn.Conv2d(1, 1, (3, 3), 1)  # in_channel , out_channel , kennel_size , stride
conv3.weight.data = torch.Tensor([[[[-1, -1, 1],
                                    [-1, 1, -1],
                                    [1, -1, -1]]
                                   ]])
 
feature_map1 = conv1(x)
feature_map2 = conv2(x)
feature_map3 = conv3(x)
 
print(feature_map1 / 9)
print(feature_map2 / 9)
print(feature_map3 / 9)
 
print("--------------- 池化  ---------------")
max_pool = nn.MaxPool2d(2, padding=0, stride=2)  # Pooling
zeroPad = nn.ZeroPad2d(padding=(0, 1, 0, 1))  # pad 0 , Left Right Up Down
 
feature_map_pad_0_1 = zeroPad(feature_map1)
feature_pool_1 = max_pool(feature_map_pad_0_1)
feature_map_pad_0_2 = zeroPad(feature_map2)
feature_pool_2 = max_pool(feature_map_pad_0_2)
feature_map_pad_0_3 = zeroPad(feature_map3)
feature_pool_3 = max_pool(feature_map_pad_0_3)
 
print(feature_pool_1.size())
print(feature_pool_1 / 9)
print(feature_pool_2 / 9)
print(feature_pool_3 / 9)
 
print("--------------- 激活  ---------------")
activation_function = nn.ReLU()
 
feature_relu1 = activation_function(feature_map1)
feature_relu2 = activation_function(feature_map2)
feature_relu3 = activation_function(feature_map3)
print(feature_relu1 / 9)
print(feature_relu2 / 9)
print(feature_relu3 / 9)

运行结果：

torch.Size([1, 1, 9, 9])
tensor([[[[-1., -1., -1., -1., -1., -1., -1., -1., -1.],
          [-1.,  1., -1., -1., -1., -1., -1.,  1., -1.],
          [-1., -1.,  1., -1., -1., -1.,  1., -1., -1.],
          [-1., -1., -1.,  1., -1.,  1., -1., -1., -1.],
          [-1., -1., -1., -1.,  1., -1., -1., -1., -1.],
          [-1., -1., -1.,  1., -1.,  1., -1., -1., -1.],
          [-1., -1.,  1., -1., -1., -1.,  1., -1., -1.],
          [-1.,  1., -1., -1., -1., -1., -1.,  1., -1.],
          [-1., -1., -1., -1., -1., -1., -1., -1., -1.]]]])
--------------- 卷积  ---------------
tensor([[[[ 0.7624, -0.1265,  0.0958,  0.3180,  0.5402, -0.1265,  0.3180],
          [-0.1265,  0.9847, -0.1265,  0.3180, -0.1265,  0.0958, -0.1265],
          [ 0.0958, -0.1265,  0.9847, -0.3487,  0.0958, -0.1265,  0.5402],
          [ 0.3180,  0.3180, -0.3487,  0.5402, -0.3487,  0.3180,  0.3180],
          [ 0.5402, -0.1265,  0.0958, -0.3487,  0.9847, -0.1265,  0.0958],
          [-0.1265,  0.0958, -0.1265,  0.3180, -0.1265,  0.9847, -0.1265],
          [ 0.3180, -0.1265,  0.5402,  0.3180,  0.0958, -0.1265,  0.7624]]]],
       grad_fn=<DivBackward0>)
tensor([[[[ 0.3578, -0.5310,  0.1356, -0.0866,  0.1356, -0.5310,  0.3578],
          [-0.5310,  0.5801, -0.5310,  0.3578, -0.5310,  0.5801, -0.5310],
          [ 0.1356, -0.5310,  0.5801, -0.7533,  0.5801, -0.5310,  0.1356],
          [-0.0866,  0.3578, -0.7533,  1.0245, -0.7533,  0.3578, -0.0866],
          [ 0.1356, -0.5310,  0.5801, -0.7533,  0.5801, -0.5310,  0.1356],
          [-0.5310,  0.5801, -0.5310,  0.3578, -0.5310,  0.5801, -0.5310],
          [ 0.3578, -0.5310,  0.1356, -0.0866,  0.1356, -0.5310,  0.3578]]]],
       grad_fn=<DivBackward0>)
tensor([[[[ 0.3411, -0.1033,  0.5633,  0.3411,  0.1189, -0.1033,  0.7856],
          [-0.1033,  0.1189, -0.1033,  0.3411, -0.1033,  1.0078, -0.1033],
          [ 0.5633, -0.1033,  0.1189, -0.3256,  1.0078, -0.1033,  0.1189],
          [ 0.3411,  0.3411, -0.3256,  0.5633, -0.3256,  0.3411,  0.3411],
          [ 0.1189, -0.1033,  1.0078, -0.3256,  0.1189, -0.1033,  0.5633],
          [-0.1033,  1.0078, -0.1033,  0.3411, -0.1033,  0.1189, -0.1033],
          [ 0.7856, -0.1033,  0.1189,  0.3411,  0.5633, -0.1033,  0.3411]]]],
       grad_fn=<DivBackward0>)
--------------- 池化  ---------------
torch.Size([1, 1, 4, 4])
tensor([[[[0.9847, 0.3180, 0.5402, 0.3180],
          [0.3180, 0.9847, 0.3180, 0.5402],
          [0.5402, 0.3180, 0.9847, 0.0958],
          [0.3180, 0.5402, 0.0958, 0.7624]]]], grad_fn=<DivBackward0>)
tensor([[[[0.5801, 0.3578, 0.5801, 0.3578],
          [0.3578, 1.0245, 0.5801, 0.1356],
          [0.5801, 0.5801, 0.5801, 0.1356],
          [0.3578, 0.1356, 0.1356, 0.3578]]]], grad_fn=<DivBackward0>)
tensor([[[[0.3411, 0.5633, 1.0078, 0.7856],
          [0.5633, 0.5633, 1.0078, 0.3411],
          [1.0078, 1.0078, 0.1189, 0.5633],
          [0.7856, 0.3411, 0.5633, 0.3411]]]], grad_fn=<DivBackward0>)
--------------- 激活  ---------------
tensor([[[[0.7624, 0.0000, 0.0958, 0.3180, 0.5402, 0.0000, 0.3180],
          [0.0000, 0.9847, 0.0000, 0.3180, 0.0000, 0.0958, 0.0000],
          [0.0958, 0.0000, 0.9847, 0.0000, 0.0958, 0.0000, 0.5402],
          [0.3180, 0.3180, 0.0000, 0.5402, 0.0000, 0.3180, 0.3180],
          [0.5402, 0.0000, 0.0958, 0.0000, 0.9847, 0.0000, 0.0958],
          [0.0000, 0.0958, 0.0000, 0.3180, 0.0000, 0.9847, 0.0000],
          [0.3180, 0.0000, 0.5402, 0.3180, 0.0958, 0.0000, 0.7624]]]],
       grad_fn=<DivBackward0>)
tensor([[[[0.3578, 0.0000, 0.1356, 0.0000, 0.1356, 0.0000, 0.3578],
          [0.0000, 0.5801, 0.0000, 0.3578, 0.0000, 0.5801, 0.0000],
          [0.1356, 0.0000, 0.5801, 0.0000, 0.5801, 0.0000, 0.1356],
          [0.0000, 0.3578, 0.0000, 1.0245, 0.0000, 0.3578, 0.0000],
          [0.1356, 0.0000, 0.5801, 0.0000, 0.5801, 0.0000, 0.1356],
          [0.0000, 0.5801, 0.0000, 0.3578, 0.0000, 0.5801, 0.0000],
          [0.3578, 0.0000, 0.1356, 0.0000, 0.1356, 0.0000, 0.3578]]]],
       grad_fn=<DivBackward0>)
tensor([[[[0.3411, 0.0000, 0.5633, 0.3411, 0.1189, 0.0000, 0.7856],
          [0.0000, 0.1189, 0.0000, 0.3411, 0.0000, 1.0078, 0.0000],
          [0.5633, 0.0000, 0.1189, 0.0000, 1.0078, 0.0000, 0.1189],
          [0.3411, 0.3411, 0.0000, 0.5633, 0.0000, 0.3411, 0.3411],
          [0.1189, 0.0000, 1.0078, 0.0000, 0.1189, 0.0000, 0.5633],
          [0.0000, 1.0078, 0.0000, 0.3411, 0.0000, 0.1189, 0.0000],
          [0.7856, 0.0000, 0.1189, 0.3411, 0.5633, 0.0000, 0.3411]]]],
       grad_fn=<DivBackward0>)

3. 可视化：了解数字与图像之间的关系

可视化卷积核和特征图

import torch
import torch.nn as nn
import matplotlib.pyplot as plt
import os
 
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
 
plt.rcParams['font.sans-serif'] = ['SimHei']  # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False  # 用来正常显示负号 #有中文出现的情况，需要u'内容
x = torch.tensor([[[[-1, -1, -1, -1, -1, -1, -1, -1, -1],
                    [-1, 1, -1, -1, -1, -1, -1, 1, -1],
                    [-1, -1, 1, -1, -1, -1, 1, -1, -1],
                    [-1, -1, -1, 1, -1, 1, -1, -1, -1],
                    [-1, -1, -1, -1, 1, -1, -1, -1, -1],
                    [-1, -1, -1, 1, -1, 1, -1, -1, -1],
                    [-1, -1, 1, -1, -1, -1, 1, -1, -1],
                    [-1, 1, -1, -1, -1, -1, -1, 1, -1],
                    [-1, -1, -1, -1, -1, -1, -1, -1, -1]]]], dtype=torch.float)
print(x.shape)
print(x)
img = x.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('原图')
plt.show()
 
print("--------------- 卷积  ---------------")
conv1 = nn.Conv2d(1, 1, (3, 3), 1)  # in_channel , out_channel , kennel_size , stride
conv1.weight.data = torch.Tensor([[[[1, -1, -1],
                                    [-1, 1, -1],
                                    [-1, -1, 1]]
                                   ]])
img = conv1.weight.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('Kernel 1')
plt.show()
conv2 = nn.Conv2d(1, 1, (3, 3), 1)  # in_channel , out_channel , kennel_size , stride
conv2.weight.data = torch.Tensor([[[[1, -1, 1],
                                    [-1, 1, -1],
                                    [1, -1, 1]]
                                   ]])
img = conv2.weight.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('Kernel 2')
plt.show()
conv3 = nn.Conv2d(1, 1, (3, 3), 1)  # in_channel , out_channel , kennel_size , stride
conv3.weight.data = torch.Tensor([[[[-1, -1, 1],
                                    [-1, 1, -1],
                                    [1, -1, -1]]
                                   ]])
img = conv3.weight.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('Kernel 3')
plt.show()
 
feature_map1 = conv1(x)
feature_map2 = conv2(x)
feature_map3 = conv3(x)
 
print(feature_map1 / 9)
print(feature_map2 / 9)
print(feature_map3 / 9)
 
img = feature_map1.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('卷积后的特征图1')
plt.show()
 
print("--------------- 池化  ---------------")
max_pool = nn.MaxPool2d(2, padding=0, stride=2)  # Pooling
zeroPad = nn.ZeroPad2d(padding=(0, 1, 0, 1))  # pad 0 , Left Right Up Down
 
feature_map_pad_0_1 = zeroPad(feature_map1)
feature_pool_1 = max_pool(feature_map_pad_0_1)
feature_map_pad_0_2 = zeroPad(feature_map2)
feature_pool_2 = max_pool(feature_map_pad_0_2)
feature_map_pad_0_3 = zeroPad(feature_map3)
feature_pool_3 = max_pool(feature_map_pad_0_3)
 
print(feature_pool_1.size())
print(feature_pool_1 / 9)
print(feature_pool_2 / 9)
print(feature_pool_3 / 9)
img = feature_pool_1.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('卷积池化后的特征图1')
plt.show()
 
print("--------------- 激活  ---------------")
activation_function = nn.ReLU()
 
feature_relu1 = activation_function(feature_map1)
feature_relu2 = activation_function(feature_map2)
feature_relu3 = activation_function(feature_map3)
print(feature_relu1 / 9)
print(feature_relu2 / 9)
print(feature_relu3 / 9)
img = feature_relu1.data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.title('卷积 + relu 后的特征图1')
plt.show()

运行结果：

卷积核

 

 

特征图：

 

 

torch.Size([1, 1, 9, 9])
tensor([[[[-1., -1., -1., -1., -1., -1., -1., -1., -1.],
          [-1.,  1., -1., -1., -1., -1., -1.,  1., -1.],
          [-1., -1.,  1., -1., -1., -1.,  1., -1., -1.],
          [-1., -1., -1.,  1., -1.,  1., -1., -1., -1.],
          [-1., -1., -1., -1.,  1., -1., -1., -1., -1.],
          [-1., -1., -1.,  1., -1.,  1., -1., -1., -1.],
          [-1., -1.,  1., -1., -1., -1.,  1., -1., -1.],
          [-1.,  1., -1., -1., -1., -1., -1.,  1., -1.],
          [-1., -1., -1., -1., -1., -1., -1., -1., -1.]]]])
--------------- 卷积  ---------------
tensor([[[[ 0.7988, -0.0901,  0.1322,  0.3544,  0.5766, -0.0901,  0.3544],
          [-0.0901,  1.0210, -0.0901,  0.3544, -0.0901,  0.1322, -0.0901],
          [ 0.1322, -0.0901,  1.0210, -0.3123,  0.1322, -0.0901,  0.5766],
          [ 0.3544,  0.3544, -0.3123,  0.5766, -0.3123,  0.3544,  0.3544],
          [ 0.5766, -0.0901,  0.1322, -0.3123,  1.0210, -0.0901,  0.1322],
          [-0.0901,  0.1322, -0.0901,  0.3544, -0.0901,  1.0210, -0.0901],
          [ 0.3544, -0.0901,  0.5766,  0.3544,  0.1322, -0.0901,  0.7988]]]],
       grad_fn=<DivBackward0>)
tensor([[[[ 0.3513, -0.5376,  0.1291, -0.0931,  0.1291, -0.5376,  0.3513],
          [-0.5376,  0.5735, -0.5376,  0.3513, -0.5376,  0.5735, -0.5376],
          [ 0.1291, -0.5376,  0.5735, -0.7598,  0.5735, -0.5376,  0.1291],
          [-0.0931,  0.3513, -0.7598,  1.0180, -0.7598,  0.3513, -0.0931],
          [ 0.1291, -0.5376,  0.5735, -0.7598,  0.5735, -0.5376,  0.1291],
          [-0.5376,  0.5735, -0.5376,  0.3513, -0.5376,  0.5735, -0.5376],
          [ 0.3513, -0.5376,  0.1291, -0.0931,  0.1291, -0.5376,  0.3513]]]],
       grad_fn=<DivBackward0>)
tensor([[[[ 0.2998, -0.1446,  0.5220,  0.2998,  0.0776, -0.1446,  0.7443],
          [-0.1446,  0.0776, -0.1446,  0.2998, -0.1446,  0.9665, -0.1446],
          [ 0.5220, -0.1446,  0.0776, -0.3669,  0.9665, -0.1446,  0.0776],
          [ 0.2998,  0.2998, -0.3669,  0.5220, -0.3669,  0.2998,  0.2998],
          [ 0.0776, -0.1446,  0.9665, -0.3669,  0.0776, -0.1446,  0.5220],
          [-0.1446,  0.9665, -0.1446,  0.2998, -0.1446,  0.0776, -0.1446],
          [ 0.7443, -0.1446,  0.0776,  0.2998,  0.5220, -0.1446,  0.2998]]]],
       grad_fn=<DivBackward0>)
--------------- 池化  ---------------
torch.Size([1, 1, 4, 4])
tensor([[[[1.0210, 0.3544, 0.5766, 0.3544],
          [0.3544, 1.0210, 0.3544, 0.5766],
          [0.5766, 0.3544, 1.0210, 0.1322],
          [0.3544, 0.5766, 0.1322, 0.7988]]]], grad_fn=<DivBackward0>)
tensor([[[[0.5735, 0.3513, 0.5735, 0.3513],
          [0.3513, 1.0180, 0.5735, 0.1291],
          [0.5735, 0.5735, 0.5735, 0.1291],
          [0.3513, 0.1291, 0.1291, 0.3513]]]], grad_fn=<DivBackward0>)
tensor([[[[0.2998, 0.5220, 0.9665, 0.7443],
          [0.5220, 0.5220, 0.9665, 0.2998],
          [0.9665, 0.9665, 0.0776, 0.5220],
          [0.7443, 0.2998, 0.5220, 0.2998]]]], grad_fn=<DivBackward0>)
--------------- 激活  ---------------
tensor([[[[0.7988, 0.0000, 0.1322, 0.3544, 0.5766, 0.0000, 0.3544],
          [0.0000, 1.0210, 0.0000, 0.3544, 0.0000, 0.1322, 0.0000],
          [0.1322, 0.0000, 1.0210, 0.0000, 0.1322, 0.0000, 0.5766],
          [0.3544, 0.3544, 0.0000, 0.5766, 0.0000, 0.3544, 0.3544],
          [0.5766, 0.0000, 0.1322, 0.0000, 1.0210, 0.0000, 0.1322],
          [0.0000, 0.1322, 0.0000, 0.3544, 0.0000, 1.0210, 0.0000],
          [0.3544, 0.0000, 0.5766, 0.3544, 0.1322, 0.0000, 0.7988]]]],
       grad_fn=<DivBackward0>)
tensor([[[[0.3513, 0.0000, 0.1291, 0.0000, 0.1291, 0.0000, 0.3513],
          [0.0000, 0.5735, 0.0000, 0.3513, 0.0000, 0.5735, 0.0000],
          [0.1291, 0.0000, 0.5735, 0.0000, 0.5735, 0.0000, 0.1291],
          [0.0000, 0.3513, 0.0000, 1.0180, 0.0000, 0.3513, 0.0000],
          [0.1291, 0.0000, 0.5735, 0.0000, 0.5735, 0.0000, 0.1291],
          [0.0000, 0.5735, 0.0000, 0.3513, 0.0000, 0.5735, 0.0000],
          [0.3513, 0.0000, 0.1291, 0.0000, 0.1291, 0.0000, 0.3513]]]],
       grad_fn=<DivBackward0>)
tensor([[[[0.2998, 0.0000, 0.5220, 0.2998, 0.0776, 0.0000, 0.7443],
          [0.0000, 0.0776, 0.0000, 0.2998, 0.0000, 0.9665, 0.0000],
          [0.5220, 0.0000, 0.0776, 0.0000, 0.9665, 0.0000, 0.0776],
          [0.2998, 0.2998, 0.0000, 0.5220, 0.0000, 0.2998, 0.2998],
          [0.0776, 0.0000, 0.9665, 0.0000, 0.0776, 0.0000, 0.5220],
          [0.0000, 0.9665, 0.0000, 0.2998, 0.0000, 0.0776, 0.0000],
          [0.7443, 0.0000, 0.0776, 0.2998, 0.5220, 0.0000, 0.2998]]]],
       grad_fn=<DivBackward0>)

二、 基于CNN的XO识别

1.数据集

共2000张图片，X、O各1000张。

从X、O文件夹，分别取出150张作为测试集。

文件夹train_data：放置训练集 1700张图片

文件夹test_data： 放置测试集 300张图片

 2.构建模型

import torch
import torch.nn as nn
 
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 9, 3)
        self.maxpool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(9, 5, 3)
 
        self.relu = nn.ReLU()
        self.fc1 = nn.Linear(27 * 27 * 5, 1200)
        self.fc2 = nn.Linear(1200, 64)
        self.fc3 = nn.Linear(64, 2)
 
    def forward(self, x):
        x = self.maxpool(self.relu(self.conv1(x)))
        x = self.maxpool(self.relu(self.conv2(x)))
        x = x.view(-1, 27 * 27 * 5)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x

3.训练模型

import torch
import torch.nn as nn
import matplotlib.pyplot as plt
from torchvision import transforms, datasets
from torch.utils.data import DataLoader
import torch.optim as optim
 
transforms = transforms.Compose([
    transforms.ToTensor(),  # 把图片进行归一化，并把数据转换成Tensor类型
    transforms.Grayscale(1)  # 把图片 转为灰度图
])
 
path = r'D:\Desktop\本学期(实验)作业\神经网络与深度学习\data\test_data'
path_test = r'D:\Desktop\本学期(实验)作业\神经网络与深度学习\data\train_data'
 
data_train = datasets.ImageFolder(path, transform=transforms)
data_test = datasets.ImageFolder(path_test, transform=transforms)
 
print("size of train_data:",len(data_train))
print("size of test_data:",len(data_test))
 
data_loader = DataLoader(data_train, batch_size=64, shuffle=True)
data_loader_test = DataLoader(data_test, batch_size=64, shuffle=True)
model = Net()
 
criterion = torch.nn.CrossEntropyLoss()  # 损失函数 交叉熵损失函数
optimizer = optim.SGD(model.parameters(), lr=0.1)  # 优化函数：随机梯度下降
 
epochs = 10
for epoch in range(epochs):
    running_loss = 0.0
    for i, data in enumerate(data_loader):
        images, label = data
        out = model(images)
        loss = criterion(out, label)
 
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
 
        running_loss += loss.item()
        if (i + 1) % 10 == 0:
            print('[%d  %5d]   loss: %.3f' % (epoch + 1, i + 1, running_loss / 100))
            running_loss = 0.0
 
print('finished train')
 
# 保存模型
torch.save(model, 'model_name.pth')  # 保存的是模型， 不止是w和b权重值

 运行结果：

4. 测试训练好的模型

# 读取模型
model_load = torch.load('model_name.pth')
# 读取一张图片 images[0]，测试
print("label[0] truth:\t", label[0])
x = images[0]
x = x.reshape([1,1,116,116])
predicted = torch.max(model_load(x), 1)
print("label[0] predict:\t", predicted.indices)
 
img = images[0].data.squeeze().numpy()  # 将输出转换为图片的格式
plt.imshow(img, cmap='gray')
plt.show()

 运行结果：

5.计算模型的准确率

# 读取模型
model_load = Net()
model_load.load_state_dict(torch.load('model_name1.pth'))
correct = 0
total = 0
with torch.no_grad():  # 进行评测的时候网络不更新梯度
    for data in data_loader_test:  # 读取测试集
        images, labels = data
        outputs = model_load(images)
        _, predicted = torch.max(outputs.data, 1)  # 取出 最大值的索引 作为 分类结果
        total += labels.size(0)  # labels 的长度
        correct += (predicted == labels).sum().item()  # 预测正确的数目
print('Accuracy of the network on the  test images: %f %%' % (100. * correct / total))

6.查看训练好的模型的特征图

# 看看每层的 卷积核 长相，特征图 长相
# 获取网络结构的特征矩阵并可视化
import torch
import matplotlib.pyplot as plt
import numpy as np
from torchvision import transforms, datasets
import torch.nn as nn
from torch.utils.data import DataLoader
 
#  定义图像预处理过程(要与网络模型训练过程中的预处理过程一致)
transforms = transforms.Compose([
    transforms.ToTensor(),  # 把图片进行归一化，并把数据转换成Tensor类型
    transforms.Grayscale(1)  # 把图片 转为灰度图
])
path = r'D:\Desktop\本学期(实验)作业\神经网络与深度学习\data\train_data'
data_train = datasets.ImageFolder(path, transform=transforms)
data_loader = DataLoader(data_train, batch_size=64, shuffle=True)
for i, data in enumerate(data_loader):
    images, labels = data
    print(images.shape)
    print(labels.shape)
    break
 
 
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 9, 3)  # in_channel , out_channel , kennel_size , stride
        self.maxpool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(9, 5, 3)  # in_channel , out_channel , kennel_size , stride
 
        self.relu = nn.ReLU()
        self.fc1 = nn.Linear(27 * 27 * 5, 1200)  # full connect 1
        self.fc2 = nn.Linear(1200, 64)  # full connect 2
        self.fc3 = nn.Linear(64, 2)  # full connect 3
 
    def forward(self, x):
        outputs = []
        x = self.conv1(x)
        outputs.append(x)
        x = self.relu(x)
        outputs.append(x)
        x = self.maxpool(x)
        outputs.append(x)
        x = self.conv2(x)
        x = self.relu(x)
        x = self.maxpool(x)
        x = x.view(-1, 27 * 27 * 5)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return outputs
 
 
# create model
model1 = Net()
 
# load model weights加载预训练权重
model_weight_path = "model_name1.pth"
model1.load_state_dict(torch.load(model_weight_path))
 
# 打印出模型的结构
print(model1)
x = images[0]
x = x.reshape([1, x.shape[0], x.shape[1], x.shape[2]])
# forward正向传播过程
out_put = model1(x)
for feature_map in out_put:
    im = np.squeeze(feature_map.detach().numpy())
    im = np.transpose(im, [1, 2, 0])
    print(im.shape)
 
    plt.figure()
    for i in range(9):
        ax = plt.subplot(3, 3, i + 1)
        plt.imshow(im[:, :, i], cmap='gray')
    plt.show()

运行结果：

 

 

7.查看训练好的模型的卷积核

# 看看每层的 卷积核 长相，特征图 长相
# 获取网络结构的特征矩阵并可视化
import torch
import matplotlib.pyplot as plt
import numpy as np
from PIL import Image
from torchvision import transforms, datasets
import torch.nn as nn
from torch.utils.data import DataLoader
 
plt.rcParams['font.sans-serif'] = ['SimHei']  # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False  # 用来正常显示负号 #有中文出现的情况，需要u'内容
#  定义图像预处理过程(要与网络模型训练过程中的预处理过程一致)
transforms = transforms.Compose([
    transforms.ToTensor(),  # 把图片进行归一化，并把数据转换成Tensor类型
    transforms.Grayscale(1)  # 把图片 转为灰度图
])
path = r'training_data_sm'
data_train = datasets.ImageFolder(path, transform=transforms)
data_loader = DataLoader(data_train, batch_size=64, shuffle=True)
for i, data in enumerate(data_loader):
    images, labels = data
    # print(images.shape)
    # print(labels.shape)
    break
 
 
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 9, 3)  # in_channel , out_channel , kennel_size , stride
        self.maxpool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(9, 5, 3)  # in_channel , out_channel , kennel_size , stride
 
        self.relu = nn.ReLU()
        self.fc1 = nn.Linear(27 * 27 * 5, 1200)  # full connect 1
        self.fc2 = nn.Linear(1200, 64)  # full connect 2
        self.fc3 = nn.Linear(64, 2)  # full connect 3
 
    def forward(self, x):
        outputs = []
        x = self.maxpool(self.relu(self.conv1(x)))
        # outputs.append(x)
        x = self.maxpool(self.relu(self.conv2(x)))
        outputs.append(x)
        x = x.view(-1, 27 * 27 * 5)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return outputs
 
 
# create model
model1 = Net()
 
# load model weights加载预训练权重
model_weight_path = "model_name1.pth"
model1.load_state_dict(torch.load(model_weight_path))
 
x = images[0]
 
# forward正向传播过程
out_put = model1(x)
 
weights_keys = model1.state_dict().keys()
for key in weights_keys:
    print("key :", key)
    # 卷积核通道排列顺序 [kernel_number, kernel_channel, kernel_height, kernel_width]
    if key == "conv1.weight":
        weight_t = model1.state_dict()[key].numpy()
        print("weight_t.shape", weight_t.shape)
        k = weight_t[:, 0, :, :]  # 获取第一个卷积核的信息参数
        # show 9 kernel ,1 channel
        plt.figure()
 
        for i in range(9):
            ax = plt.subplot(3, 3, i + 1)  # 参数意义：3：图片绘制行数，5：绘制图片列数，i+1：图的索引
            plt.imshow(k[i, :, :], cmap='gray')
            title_name = 'kernel' + str(i) + ',channel1'
            plt.title(title_name)
        plt.show()
 
    if key == "conv2.weight":
        weight_t = model1.state_dict()[key].numpy()
        print("weight_t.shape", weight_t.shape)
        k = weight_t[:, :, :, :]  # 获取第一个卷积核的信息参数
        print(k.shape)
        print(k)
 
        plt.figure()
        for c in range(9):
            channel = k[:, c, :, :]
            for i in range(5):
                ax = plt.subplot(2, 3, i + 1)  # 参数意义：3：图片绘制行数，5：绘制图片列数，i+1：图的索引
                plt.imshow(channel[i, :, :], cmap='gray')
                title_name = 'kernel' + str(i) + ',channel' + str(c)
                plt.title(title_name)
            plt.show()

 运行结果：

 

 

 

 

 

 

 

 

 

key : conv1.weight
weight_t.shape (9, 1, 3, 3)
key : conv1.bias
key : conv2.weight
weight_t.shape (5, 9, 3, 3)
(5, 9, 3, 3)
[[[[ 6.98458925e-02  5.90659641e-02  3.81753966e-02]
   [-6.13637008e-02  1.34813562e-01 -7.33171105e-02]
   [ 6.13309480e-02 -5.23449155e-04  4.80439179e-02]]

  [[-4.00464647e-02  1.12880126e-01 -7.68649280e-02]
   [ 7.87192211e-02 -2.36817487e-02 -5.60271740e-02]
   [ 8.12683925e-02  2.73875427e-02 -9.82172340e-02]]

  [[ 4.08881977e-02  5.25458008e-02  1.05747022e-03]
   [ 1.61746703e-02 -1.88982282e-02 -8.98489915e-03]
   [ 2.05815379e-02  6.91843778e-02 -5.07958978e-02]]

  [[ 2.40698364e-02  1.23903854e-02  9.09792781e-02]
   [-4.54008244e-02  1.94115769e-02  1.08469449e-01]
   [-5.29113822e-02  5.41664548e-02  4.23667813e-03]]

  [[-9.22499970e-02 -2.69414317e-02  5.68897910e-02]
   [-9.98499840e-02  6.78706095e-02 -7.92549327e-02]
   [-4.87312488e-02 -5.90884238e-02  7.44135678e-02]]

  [[-8.56655538e-02 -8.28199610e-02  8.21587667e-02]
   [ 2.01716665e-02  8.60504657e-02 -4.26697880e-02]
   [-4.74872179e-02  1.12074219e-01 -5.09875007e-02]]

  [[ 3.27787511e-02 -7.82772973e-02  1.08398296e-01]
   [ 2.77021844e-02  3.72366756e-02 -6.03387356e-02]
   [ 3.33355628e-02 -3.38297561e-02  1.04788266e-01]]

  [[-5.93599454e-02  7.44077489e-02  1.03815481e-01]
   [-1.66368708e-02 -9.41569544e-03  3.45407762e-02]
   [ 3.65562551e-03 -4.04765494e-02  6.14924431e-02]]

  [[ 1.00850463e-01  3.62970792e-02  1.14644319e-01]
   [-9.65513289e-02 -2.99563049e-03  9.44290385e-02]
   [-5.09380549e-02  8.94127786e-02  8.03329796e-02]]]


 [[[ 2.16829151e-01  1.99093044e-01  5.70072345e-02]
   [ 2.30329320e-01  4.84957658e-02 -2.16806829e-02]
   [ 2.35382617e-01  1.77218691e-01  1.56592391e-02]]

  [[-1.14416555e-01 -9.05001387e-02  3.45594026e-02]
   [ 7.52960369e-02  4.63159829e-02 -1.27104968e-01]
   [-6.35510609e-02  9.67403129e-02 -2.61216182e-02]]

  [[ 5.76992482e-02 -4.30453755e-03 -4.52346914e-02]
   [-9.95732378e-03  9.97230113e-02 -7.94924702e-03]
   [ 2.94310488e-02 -6.90363571e-02 -6.99906722e-02]]

  [[ 1.05409855e-02  1.29709765e-02  3.56001705e-02]
   [-1.83131136e-02  1.66197456e-02 -1.48420064e-02]
   [ 7.52180666e-02 -7.81482905e-02 -9.23441350e-03]]

  [[-9.07110646e-02 -2.30383456e-01 -2.82495804e-02]
   [-1.90353051e-01 -4.77160737e-02  5.00900671e-02]
   [-1.54941514e-01 -1.49308849e-04  2.56049205e-02]]

  [[ 4.10602428e-03 -9.44532547e-03  5.52874431e-02]
   [ 3.63270789e-02  2.35939752e-02 -6.39206991e-02]
   [ 1.41671211e-01  7.78410360e-02 -8.86475742e-02]]

  [[-4.53491695e-02  1.07495286e-01  5.71262278e-02]
   [-1.80877224e-02  1.15991592e-01 -1.03785552e-01]
   [-6.88361079e-02  3.28751318e-02  3.43425907e-02]]

  [[ 1.12147458e-01  8.87614563e-02 -6.72129467e-02]
   [-3.03914417e-02 -5.64941065e-03 -8.04529339e-03]
   [ 9.83864889e-02  1.11261480e-01 -9.43764746e-02]]

  [[ 1.46605387e-01  1.30711570e-01  3.23899859e-03]
   [ 8.96535888e-02  2.28065569e-02 -1.84486173e-02]
   [ 4.97324243e-02  1.14006259e-01 -4.68446128e-02]]]


 [[[ 7.06592649e-02  1.04841381e-01 -8.21368843e-02]
   [ 7.56961759e-03  3.07901427e-02  1.26209920e-02]
   [-4.03140374e-02  1.02723790e-02 -3.61617506e-02]]

  [[-1.27743632e-01  7.13284835e-02 -1.00524239e-02]
   [ 8.56978819e-03 -8.83464962e-02 -1.46895722e-01]
   [-9.48935673e-02 -3.92020792e-02 -5.77922463e-02]]

  [[ 6.93405345e-02 -6.33155406e-02 -1.07146412e-01]
   [-2.40081158e-02  1.51393469e-02 -8.50542560e-02]
   [ 1.03771903e-01  3.33332270e-02 -3.12288590e-02]]

  [[ 8.57113898e-02 -5.51949926e-02  1.79728121e-02]
   [ 6.85685426e-02  7.82678053e-02 -2.39495300e-02]
   [-5.95061481e-02 -1.02192089e-01 -9.46415141e-02]]

  [[-1.26110375e-01 -2.46208310e-02 -1.84309453e-01]
   [ 1.56330317e-02 -1.09103754e-01  7.10281078e-03]
   [ 2.81431363e-03  2.14111097e-02 -6.03581294e-02]]

  [[-3.36449742e-02 -1.04325727e-01  4.81084827e-03]
   [ 3.85972895e-02 -1.06708854e-01  1.05884507e-01]
   [-5.22216298e-02 -1.23345144e-02  1.25243125e-04]]

  [[ 6.98298663e-02 -7.50838518e-02  2.16020737e-02]
   [ 2.92224884e-02 -2.44561583e-03  9.43801999e-02]
   [-2.86634099e-02 -3.26344781e-02  2.32272912e-02]]

  [[ 1.64350141e-02  4.88247387e-02  6.04826063e-02]
   [-2.23416858e-03  2.94138119e-02  7.57313818e-02]
   [-7.45175313e-03 -8.91718641e-02  1.07973032e-01]]

  [[-5.71223162e-02 -1.00419804e-01 -9.37241390e-02]
   [-6.41234219e-02  1.57135073e-02 -1.89927071e-02]
   [ 3.19132023e-02  3.52603830e-02 -5.65539226e-02]]]


 [[[-7.09892437e-02 -7.83143938e-02 -6.47149161e-02]
   [-3.67178880e-02 -7.28780478e-02  4.98269778e-03]
   [-8.59382451e-02  7.90102500e-03  8.22086632e-02]]

  [[-9.37657654e-02  2.11495031e-02 -9.26158875e-02]
   [-4.10797484e-02 -3.01056765e-02 -1.08754538e-01]
   [-1.06991671e-01 -9.82748047e-02 -1.05811022e-01]]

  [[ 1.04221590e-01  6.28329888e-02 -3.75137329e-02]
   [ 3.30754891e-02  3.89316007e-02 -7.92906582e-02]
   [ 7.55222887e-03  5.62973954e-02 -1.11638568e-02]]

  [[ 8.80549848e-03 -6.03972003e-02  4.41536307e-03]
   [-3.37611958e-02  1.01789005e-01 -7.89848119e-02]
   [-9.12546590e-02  1.03225388e-01 -9.06298384e-02]]

  [[-1.10043190e-01 -2.76146475e-02 -5.13514876e-02]
   [-5.09044081e-02  1.09844722e-01  4.63054404e-02]
   [ 1.08196847e-01 -9.77096856e-02 -6.03941232e-02]]

  [[ 5.21519445e-02 -9.87846181e-02 -8.58242884e-02]
   [ 7.53185526e-02 -1.06183343e-01 -2.14587171e-02]
   [-1.01366051e-01  1.07559510e-01 -7.94448555e-02]]

  [[ 6.74566850e-02  9.86777246e-04  5.59085682e-02]
   [ 2.23980322e-02  2.54910961e-02 -9.46119949e-02]
   [ 1.42240748e-02  8.70785415e-02  1.08000107e-01]]

  [[ 3.61804152e-03 -4.88866568e-02 -6.17030747e-02]
   [-9.81834158e-02 -3.27483080e-02  8.97164792e-02]
   [-8.79103914e-02  4.35663499e-02 -5.99438958e-02]]

  [[-3.83332074e-02  9.47101191e-02 -1.09892726e-01]
   [-7.83142447e-03 -8.67927223e-02  6.75773546e-02]
   [-1.04694366e-02 -5.16315363e-02  9.21809003e-02]]]


 [[[ 1.57988280e-01  3.60716939e-01  2.89625138e-01]
   [ 3.33296895e-01  3.04307610e-01  2.04062670e-01]
   [ 3.17273408e-01  2.26779416e-01  2.79705107e-01]]

  [[ 1.01867160e-02 -7.60785118e-02  6.69518635e-02]
   [ 5.54578081e-02  2.87681669e-02 -4.60735569e-03]
   [ 1.03337668e-01 -1.09475963e-02 -3.04092001e-02]]

  [[-9.26796421e-02 -4.36972715e-02  2.43944284e-02]
   [ 5.30454889e-02  2.84075048e-02  1.89190935e-02]
   [ 8.29062089e-02 -4.56020273e-02 -4.39854078e-02]]

  [[-7.51409829e-02  7.07833990e-02 -2.27690194e-04]
   [ 5.10067791e-02  2.99685933e-02 -6.13842085e-02]
   [-6.48679212e-03  1.06581427e-01  4.64117266e-02]]

  [[-8.37772042e-02 -1.55725271e-01 -1.20179079e-01]
   [ 1.86880562e-03  1.55261280e-02 -1.57763153e-01]
   [-4.79777083e-02  2.06353310e-02 -1.45830363e-01]]

  [[ 2.58288588e-02  2.36822367e-02  5.96350171e-02]
   [-2.38394365e-02  2.01987803e-01  7.84047414e-03]
   [ 6.91130161e-02  1.77542195e-01  7.41390362e-02]]

  [[-3.30275223e-02  9.67713445e-02 -8.22414979e-02]
   [-2.45989729e-02  1.26603991e-01 -5.13266437e-02]
   [ 1.19841211e-01  8.46858248e-02 -3.77128050e-02]]

  [[ 1.54069349e-01  1.45480275e-01  9.25109535e-03]
   [ 1.04882568e-01  1.83798082e-03  6.50073215e-02]
   [ 6.25121519e-02  1.17242105e-01 -2.45001912e-03]]

  [[ 1.30424738e-01  1.92920014e-01  5.10507822e-03]
   [ 7.14448616e-02  1.48681432e-01 -2.15764381e-02]
   [ 1.92923009e-01  1.89543799e-01 -3.10594495e-02]]]]
key : conv2.bias
key : fc1.weight
key : fc1.bias
key : fc2.weight
key : fc2.bias
key : fc3.weight
key : fc3.bias

 8.训练模型源代码

import torch
from torchvision import transforms, datasets
import torch.nn as nn
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
import torch.optim as optim
 
transforms = transforms.Compose([
    transforms.ToTensor(),  # 把图片进行归一化，并把数据转换成Tensor类型
    transforms.Grayscale(1)  # 把图片 转为灰度图
])
 
path = r'train_data'
path_test = r'test_data'
 
data_train = datasets.ImageFolder(path, transform=transforms)
data_test = datasets.ImageFolder(path_test, transform=transforms)
 
print("size of train_data:",len(data_train))
print("size of test_data:",len(data_test))
 
data_loader = DataLoader(data_train, batch_size=64, shuffle=True)
data_loader_test = DataLoader(data_test, batch_size=64, shuffle=True)
 
for i, data in enumerate(data_loader):
    images, labels = data
    print(images.shape)
    print(labels.shape)
    break
 
for i, data in enumerate(data_loader_test):
    images, labels = data
    print(images.shape)
    print(labels.shape)
    break
 
 
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 9, 3)  # in_channel , out_channel , kennel_size , stride
        self.maxpool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(9, 5, 3)  # in_channel , out_channel , kennel_size , stride
 
        self.relu = nn.ReLU()
        self.fc1 = nn.Linear(27 * 27 * 5, 1200)  # full connect 1
        self.fc2 = nn.Linear(1200, 64)  # full connect 2
        self.fc3 = nn.Linear(64, 2)  # full connect 3
 
    def forward(self, x):
        x = self.maxpool(self.relu(self.conv1(x)))
        x = self.maxpool(self.relu(self.conv2(x)))
        x = x.view(-1, 27 * 27 * 5)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x
 
 
model = Net()
 
criterion = torch.nn.CrossEntropyLoss()  # 损失函数 交叉熵损失函数
optimizer = optim.SGD(model.parameters(), lr=0.1)  # 优化函数：随机梯度下降
 
epochs = 10
for epoch in range(epochs):
    running_loss = 0.0
    for i, data in enumerate(data_loader):
        images, label = data
        out = model(images)
        loss = criterion(out, label)
 
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
 
        running_loss += loss.item()
        if (i + 1) % 10 == 0:
            print('[%d  %5d]   loss: %.3f' % (epoch + 1, i + 1, running_loss / 100))
            running_loss = 0.0
 
print('finished train')
 
# 保存模型 torch.save(model.state_dict(), model_path)
torch.save(model.state_dict(), 'model_name1.pth')  # 保存的是模型， 不止是w和b权重值
 
# 读取模型
model = torch.load('model_name1.pth')

9.测试模型源代码

# https://blog.csdn.net/qq_53345829/article/details/124308515
import torch
from torchvision import transforms, datasets
import torch.nn as nn
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
import torch.optim as optim
 
transforms = transforms.Compose([
    transforms.ToTensor(),  # 把图片进行归一化，并把数据转换成Tensor类型
    transforms.Grayscale(1)  # 把图片 转为灰度图
])
 
path = r'train_data'
path_test = r'test_data'
 
data_train = datasets.ImageFolder(path, transform=transforms)
data_test = datasets.ImageFolder(path_test, transform=transforms)
 
print("size of train_data:", len(data_train))
print("size of test_data:", len(data_test))
 
data_loader = DataLoader(data_train, batch_size=64, shuffle=True)
data_loader_test = DataLoader(data_test, batch_size=64, shuffle=True)
print(len(data_loader))
print(len(data_loader_test))
 
 
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 9, 3)  # in_channel , out_channel , kennel_size , stride
        self.maxpool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(9, 5, 3)  # in_channel , out_channel , kennel_size , stride
 
        self.relu = nn.ReLU()
        self.fc1 = nn.Linear(27 * 27 * 5, 1200)  # full connect 1
        self.fc2 = nn.Linear(1200, 64)  # full connect 2
        self.fc3 = nn.Linear(64, 2)  # full connect 3
 
    def forward(self, x):
        x = self.maxpool(self.relu(self.conv1(x)))
        x = self.maxpool(self.relu(self.conv2(x)))
        x = x.view(-1, 27 * 27 * 5)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x
 
# 读取模型
model = Net()
model.load_state_dict(torch.load('model_name1.pth', map_location='cpu')) # 导入网络的参数
 
# model_load = torch.load('model_name1.pth')
# https://blog.csdn.net/qq_41360787/article/details/104332706
 
correct = 0
total = 0
with torch.no_grad():  # 进行评测的时候网络不更新梯度
    for data in data_loader_test:  # 读取测试集
        images, labels = data
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)  # 取出 最大值的索引 作为 分类结果
        total += labels.size(0)  # labels 的长度
        correct += (predicted == labels).sum().item()  # 预测正确的数目
print('Accuracy of the network on the  test images: %f %%' % (100. * correct / total))
 
#  "_," 的解释 https://blog.csdn.net/weixin_48249563/article/details/111387501

总结

本次实验实现了图片数据集的识别，自己区分了训练集与测试集，自己完整进行了测试过程，并进行了可视化，能够清晰看出图片卷积后的变化，卷积是课程重点，应该花更多时间进行学习。

参考

【2021-2022 春学期】人工智能-作业6：CNN实现XO识别_HBU_David的博客-CSDN博客

Stacey.933
关注
  • 1
    点赞
  • 踩
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
NNDL-solutions:迈克尔·尼尔森（Michael Nielsen）的《神经网络与深度学习》一书中的练习和问题的解决方案
05-03
NNDL解决方案 迈克尔·尼尔森（Michael Nielsen）的书《 （以及对Python 3和Theano 1.0.3的代码的改编）中的练习和问题的解决方案（数学和代码）。 在这里找到练习和问题的解决方案： 涉及数学： notebooks 涉及...
NNDL:神经网络和深度学习课程回购
04-12
作业6： 在房屋价格数据集上实现小批量梯度下降。作业7： 它由批量梯度下降优化器的实现组成，该优化器用于对点样本进行采样，这些点取自线y = 2 * x + 3的方程式。作业8： 在压力和重量数据集上实现微型批次梯度...
深度学习实用代码块集锦
MJy65535的博客
03-21 867
本文为学习深度学习过程中实用代码总结、小经验小技巧总结，持续更新~
神经网络与深度学习 作业6：基于CNN的XO识别
weixin_53651790的博客
10-22 343
基于CNN的XO识别
深度学习作业-图像识别
qq_34112576的博客
03-16 2730
深度学习作业-图像识别 一、AlexNet进行中文字体的识别 1 、数据准备 2 、构造数据生成器 3 、输出图像 4 、AlexNet代码实现 1 .模型结构搭建 2 .模型编译 3 .模型运行与拟合 二、自搭建网络进行中文字体的识别 1 、数据准备 2 、构造数据生成器 3 、自搭建模型代码实现 1 .模型结构搭建 2 .模型运行与拟合 3 .对新预测集进行模型预测 1 .使用带泄露的ReLU函数 2 .模型中不使用dropout 结论 数据背景 本次
halcon 深度学习（一）：分类
长春小佛爷的博客
01-23 3883
*设置窗体 *关闭窗体 dev_update_off () dev_close_window () *新建窗体 WindowWidth := 800 WindowHeight := 600 dev_open_window_fit_size (0, 0, WindowWidth, WindowHeight, -1, -1, WindowHandle) *字体 set_display_font (WindowHandle, 16, 'mono', 'true', 'false') *打上标签 *选...
Halcon 深度学习--手写数字分类（亲测）
panda HAHA的博客
03-02 1356
Halcon 版的深度学习，操作简单。
NNDL:使用python的梯度下降算法
03-25
NNDL:使用python的梯度下降算法
锻炼：锻炼nndl
02-26
5.卷积神经网络卷积神经网络（CNN） 利用卷积神经网络，处理MNIST数据集分类问题。 6.循环神经网络 基于循环神经网络的唐诗生成问题 7.注意力机制 使用序列对模型进行建模。 使用注意序列对模型进行排序。 11.高斯...
NNDL:神经网络与深度学习
03-07
NNDL:神经网络与深度学习
神经网络与深度学习 作业11：优化算法比较
weixin_53651790的博客
12-04 371
作业11：优化算法比较
【深度学习-吴恩达】L1-4 深层神经网络 作业
u011315681的博客
07-28 615
【深度学习-吴恩达】L1-4 深层神经网络 作业笔记
Halcon 深度学习（三）：缺陷检测
qq_41373415的博客
04-21 1万+
目录前言一、准备1、准备缺陷检测所需图片1.1、images文件夹：3种类型图片1.2、divisionImages文件夹：3种类型图片（用于分割背景）二、编写代码1、设置输入输出路径2、设置参数3、处理图片、拆分4、预处理数据集5、测试 前言 最近学习深度学习时的一些总结和看法，参照深度学习的自带案例（segment_pill_deep_learning_1_preprocess.hdev） 一、准备 1、准备缺陷检测所需图片 1.1、images文件夹：3种类型图片 每个类型中分为3种：conta
【HSI】高光谱的数据集分类深度学习实战及代码理解
热门推荐
川川自溢的博客
08-11 1万+
高光谱的数据集分类深度学习实战及代码理解，基于最简单的2D卷积来对整个高光谱深度学习特征分类进行讲述，模型可以进行相应的替换
Halcon学习---深度学习篇segment3~评估训练后的模型
qq_26572229的博客
09-12 1958
* *此示例是一系列示例的一部分，该示例总结了 * DL细分的工作流程。 它使用MVTec药丸数据集。 *四个部分是： * 1.数据集预处理 * 2.训练模型。 * 3.评估训练后的模型。 * 4.推断新图像。 * *此示例包含第3部分：训练模型的评估。 *请注意：此脚本需要第1部分的输出： * segment_pill_defects_deep_learning_1_preprocess.hdev *如果将UsePretrainedModel设置为：= false，则还需要第2部分的输出： * segme
halcon深度学习（分类）（1）
day_light_up的博客
10-19 2790
学习halocon自带的分类学习的例子： 创建网络和数据预处理 训练网络 评估训练的效果 测试新图像 ①设置输入输出路径：1.分类图像数据路径RawImageBaseFolder 2.设置样本数据文件夹ExampleDataDir ②设置参数：LabelSource/TraningPercent/ValidationPercent/ImageWidth/ImageHeight/ImageNumChannels/NormalizationType/DomainHanding 为了设置可再生或
HBU-NNDL 实验四 线性分类
m0_57215376的博客
09-21 452
HBU-nndl 实验四 线性分类
halcon深度学习4：深度学习在 OCR的用法-deep_ocr_workflow解析
mlxg99999的博客
08-10 7645
1. 什么是OCR技术 OCR，全称是Optical Character Recognition, 即光学字符识别，面向扫描文件。但是由于现在数字图像的普及，这里泛指文字检测和识别，包括扫描文档和自然场景的文字识别。
NNDL 作业11：优化算法比较
qq_64335800的博客
12-05 360
优化算法比较
2、编程练习：基于循环神经网络的唐诗生成问题 https://github.com/nndl/exercise/
最新发布
05-04
这个项目是基于 Tensorflow 实现的，目标是使用循环神经网络生成唐诗。 在这个项目中，你将学习如何使用 Tensorflow 实现循环神经网络，并且了解到如何在自然语言处理任务中使用它们。你将使用唐诗数据集进行模型...

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值