使用python svm实现直接可用的手写数字识别_Python

python svm实现手写数字识别——直接可用

最近在做个围棋识别的项目，需要识别下面的数字，如下图：

使用python svm实现直接可用的手写数字识别

我发现现在网上很多代码是良莠不齐，…真是一言难尽，于是记录一下，能够运行成功并识别成功的一个源码。

1、训练

1.1、训练数据集下载——已转化成csv文件

1.2 、训练源码

train.py

				?

									import pandas as pd

									from sklearn.decomposition import pca

									from sklearn import svm

									from sklearn.externals import joblib

									import time

									if __name__ =="__main__":

									    train_num = 5000

									    test_num = 7000

									    data = pd.read_csv('train.csv')

									    train_data = data.values[0:train_num,1:]

									    train_label = data.values[0:train_num,0]

									    test_data = data.values[train_num:test_num,1:]

									    test_label = data.values[train_num:test_num,0]

									    t = time.time()

									    #pca降维

									    pca = pca(n_components=0.8, whiten=true)

									    print('start pca...')

									    train_x = pca.fit_transform(train_data)

									    test_x = pca.transform(test_data)

									    print(train_x.shape)

									    # svm训练

									    print('start svc...')

									    svc = svm.svc(kernel = 'rbf', c = 10)

									    svc.fit(train_x,train_label)

									    pre = svc.predict(test_x)

									    #保存模型

									    joblib.dump(svc, 'model.m')

									    joblib.dump(pca, 'pca.m')

									    # 计算准确率

									    score = svc.score(test_x, test_label)

									    print(u'准确率：%f,花费时间：%.2fs' % (score, time.time() - t))

2、预测单张图片

2.1、待预测图像

2.2、预测源码

				?

									from sklearn.externals import joblib

									import cv2

									if __name__ =="__main__":

									    img = cv2.imread("img_temp.jpg", 0)

									    #test = img.reshape(1,1444)![在这里插入图片描述](https://img-blog.csdnimg.cn/20210630133136668.jpg#pic_center)

									    tp_x = 10

									    tp_y = 10

									    tp_width = 20

									    tp_height = 20

									    img_temp = img[tp_y:tp_y + tp_height, tp_x:tp_x + tp_width]  # 参数含义分别是：y、y+h、x、x+w

									    cv2.namedwindow("src", 0)

									    cv2.imshow("src", img_temp)

									    cv2.waitkey(1000)

									    [height, width] = img_temp.shape

									    print(width, height)

									    res_img = cv2.resize(img_temp, (28, 28))

									    test = res_img.reshape(1, 784)

									    #加载模型

									    svc = joblib.load("model.m")

									    pca = joblib.load("pca.m")

									    # svm

									    print('start pca...')

									    test_x = pca.transform(test)

									    print(test_x.shape)

									    pre = svc.predict(test_x)

									    print(pre[0])