KNN实现简单手写识别

数据集见Source/数据集/手写数字 at main · ziwenhahaha/Source (github.com)

导包

import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.neighbors import KNeighborsClassifier
import pandas as pd
import numpy as np

随便读一张图看看

img_arr = plt.imread('./digist/5/5_1.bmp')
plt.imshow(img_arr)	

查看图片形状

img_arr.shape

迭代目录，把所有图片的特征存入feature，标签存入target

feature = []
target = []
for i in range(10):
    for j in range(1,501):
        #./digist/5/5_1.bmp
        img_path = f"./digist/{i}/{i}_{j}.bmp"
        #print(imp_path)
        img_arr = plt.imread(img_path)
        feature.append(img_arr)
        target.append(i)

转一下numpy.array格式，方面后面坍塌成二维

feature = np.array(feature)
target = np.array(target)

查看形状

feature.shape

坍塌成二维

#将featrue变形成2维
feature = feature.reshape((5000,28*28))

切分数据集

#切分数据集
x_train,x_test,y_train,y_test = train_test_split(feature,target,test_size=0.1,random_state=2021)

寻找最优的超参数

#寻找最优的超参数
ks = np.arange(5,150,5)
scores = []
for k in ks:
    knn = KNeighborsClassifier(k)
    %time score = cross_val_score(knn,x_train,y_train,cv=5).mean()
    scores.append(score)
    print(k)
plt.plot(ks,scores)

scores = np.array(scores)
max_value_index = np.argmax(scores)
print('最优的k值：',ks[max_value_index])

建立模型，用最优的超参数进行模型训练

#建立模型，基于最优的超参数进行模型训练
knn = KNeighborsClassifier(5)
knn.fit(x_train,y_train)
print(knn.score(x_test,y_test))
print('模型分类结果:',knn.predict(x_test))
print('真实的分类结果:',y_test)

pd.Series(np.array(knn.predict(x_test))==np.array(y_test)).value_counts()

用训练好的模型，去检测别的图片

#将训练好的模型作用的未知的数据中
img_arr = plt.imread('./123.jpg')
plt.imshow(img_arr)

将某个数字剪下来

#将5裁剪下来
five_img_arr = img_arr[300:435,185:295]
plt.imshow(five_img_arr)

明显剪下来不符合28*28，所以需要等比例压缩

#将five_img_arr进行像素的等比例压缩（28*28），在对其进行扁平化处理即可
import scipy.ndimage as ndimage
five_img_arr_zoom = ndimage.zoom(five_img_arr,zoom=(28/135,28/110))

预测

x_train[1].shape #随便查看一个样本，发现维度是784
knn.predict([five_img_arr_zoom.reshape(784,)])

#最后结果：array([5])  预测成功，图片就是5

完整代码

#导包
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.neighbors import KNeighborsClassifier
import pandas as pd
import numpy as np

#随便读一张图看看
img_arr = plt.imread('./digist/5/5_1.bmp')
plt.imshow(img_arr)	

#查看图片形状
img_arr.shape

#迭代目录，把所有图片的特征存入feature，标签存入target
feature = []
target = []
for i in range(10):
    for j in range(1,501):
        #./digist/5/5_1.bmp
        img_path = f"./digist/{i}/{i}_{j}.bmp"
        #print(imp_path)
        img_arr = plt.imread(img_path)
        feature.append(img_arr)
        target.append(i)

#转一下numpy.array格式，方面后面坍塌成二维
feature = np.array(feature)
target = np.array(target)

#查看形状
feature.shape

#坍塌成二维
feature = feature.reshape((5000,28*28))

#切分数据集
x_train,x_test,y_train,y_test = train_test_split(feature,target,test_size=0.1,random_state=2021)

#寻找最优的超参数
ks = np.arange(5,150,5)
scores = []
for k in ks:
    knn = KNeighborsClassifier(k)
    %time score = cross_val_score(knn,x_train,y_train,cv=5).mean()
    scores.append(score)
    print(k)
plt.plot(ks,scores)

scores = np.array(scores)
max_value_index = np.argmax(scores)
print('最优的k值：',ks[max_value_index])


#建立模型，基于最优的超参数进行模型训练
knn = KNeighborsClassifier(5)
knn.fit(x_train,y_train)
print(knn.score(x_test,y_test))
print('模型分类结果:',knn.predict(x_test))
print('真实的分类结果:',y_test)

pd.Series(np.array(knn.predict(x_test))==np.array(y_test)).value_counts()


#将训练好的模型检测别的图片
img_arr = plt.imread('./123.jpg')
plt.imshow(img_arr)


#将某个数字剪下来
five_img_arr = img_arr[300:435,185:295]
plt.imshow(five_img_arr)


#明显剪下来不符合28*28，所以需要等比例压缩

#将five_img_arr进行像素的等比例压缩（28*28），在对其进行扁平化处理即可
import scipy.ndimage as ndimage
five_img_arr_zoom = ndimage.zoom(five_img_arr,zoom=(28/135,28/110))


预测

x_train[1].shape #随便查看一个样本，发现维度是784
knn.predict([five_img_arr_zoom.reshape(784,)])

#最后结果：array([5])  预测成功，图片就是5