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：

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

百度网盘：
链接：https://pan.baidu.com/s/17_XBmtQK-lSejswmQJ6YtA
提取码：jek5

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])

2.3、预测结果