图片分类-pca_svc

PCA_SVC训练

import os
from PIL import Image
import numpy as np
from sklearn.decomposition import PCA
from sklearn.svm import SVC
import time
import datetime
def load_Img(imgDir):
    lable = os.listdir(imgDir)
    #print(lable)
    OK_name=os.listdir(imgDir+'/'+lable[0])
    NG_name=os.listdir(imgDir+'/'+lable[1])
    #print(NG_name)

    label=[]
    data=[]

    for i in range(len(OK_name)):
        start = datetime.datetime.now()

        OK_path = imgDir + "/" + lable[0]+'/'+OK_name[i-1]
        OK_img = Image.open(OK_path)

        OK_img=OK_img.convert('L')
     #   print(OK_img.size)
        end = datetime.datetime.now()
        print(end - start)

        OK_img = OK_img.resize((64,64))

        OK_arr = np.asarray(OK_img, dtype="float32")

        OK_arr = OK_arr.flatten()

        data.append(OK_arr)



        label.append(lable[0])
  #      label.append(1)

    for j in range(len(NG_name)):

        NG_path = imgDir + '/' + lable[1] + '/' + NG_name[j - 1]
        NG_img = Image.open(NG_path)
        NG_img=NG_img.convert('L')
      #  print(NG_img.size)
        NG_img = NG_img.resize((64,64))

        NG_arr = np.asarray(NG_img, dtype="float32")

        NG_arr = NG_arr.flatten()

        data.append(NG_arr)

 #       label.append(0)
        label.append(lable[1])
    return label ,data
craterDir = "E:\image\Result"

label,data = load_Img(craterDir)
#print(label)
#print(data)


#将数据分割训练数据与测试数据
from sklearn.model_selection import train_test_split
# 随机采样20%的数据构建测试样本，其余作为训练样本
X_train, X_test, y_train, y_test = train_test_split(data,label , random_state=33, test_size=0.2)

# 一个参数点（PCA维数为n）的模型训练和测试，得到该参数下模型在校验集上的预测性能
def n_component_analysis(n,C,gamma, X_train, y_train, X_val, y_val):
    #start = time.time()
    start = datetime.datetime.now()
    pca = PCA(n_components=n)
    print("PCA begin with n_components: {}".format(n));
    pca.fit(X_train)

    # 在训练集和测试集降维
    X_train_pca = pca.transform(X_train)
    X_val_pca = pca.transform(X_val)

    # 利用SVC训练
    print('SVC begin')
    clf1 = SVC(C=C,gamma=gamma)
   # clf1 = SVC(C=C,kernel='rbf',gamma=gamma)
    clf1.fit(X_train_pca, y_train)

    # 返回accuracy
    accuracy = clf1.score(X_val_pca, y_val)
    end = datetime.datetime.now()
  #  end = time.time()
    print("accuracy: {},C:{},gamma:{} time elaps:{}".format(accuracy,C,gamma ,end - start))
    return accuracy

# 设置超参数（PCA维数）搜索范围
n_s = np.linspace(0.70, 0.85, num=3)
#需要调优的参数
C_s = np.logspace(4,6, 3)# logspace(a,b,N)把10的a次方到10的b次方区间分成N份
gamma_s = np.logspace(-8, -6, 3)

accuracy = []

if __name__=='__main__':
    for n in n_s:
        for i, oneC in enumerate(C_s):
            for j, gamma in enumerate(gamma_s):
                tmp = n_component_analysis(n, oneC, gamma,X_train, y_train, X_test, y_test)
                accuracy.append(tmp)

保存模型

# coding=utf-8
import os
import sys
from PIL import Image
import numpy as np
from sklearn.svm import SVC
import pickle
from sklearn.decomposition import PCA
#将数据分割训练数据与测试数据
from sklearn.model_selection import train_test_split
# 随机采样20%的数据构建测试样本，其余作为训练样本
def Gain_Img(imgDir):
    lable = os.listdir(imgDir)
    OK_name=os.listdir(imgDir+'/'+lable[0])
    NG_name=os.listdir(imgDir+'/'+lable[1])
    print(lable)
    for i in range(len(OK_name)):


        OK_path = imgDir + "/" + lable[0]+'/'+OK_name[i-1]
        OK_img = Image.open(OK_path)
        OK_img=OK_img.convert('L')
        OK_img = OK_img.resize((256,256))
        out1 = OK_img.rotate(90)  # 逆时针旋转45度
        if not os.path.exists('rotation/{}'.format(lable[0])):
            os.makedirs('rotation/{}'.format(lable[0]))
        if not os.path.exists('rotation/{}'.format(lable[1])):
            os.makedirs('rotation/{}'.format(lable[1]))
        out1.save("rotation\\{}\\{}_90_{}.bmp".format(lable[0],lable[0],i))
        OK_img.save("rotation\\{}\\{}_{}.bmp".format(lable[0],lable[0],i))

    for j in range(len(NG_name)):

        NG_path = imgDir + '/' + lable[1] + '/' + NG_name[j - 1]
        NG_img = Image.open(NG_path)
        NG_img = NG_img.convert('L')
        NG_img = NG_img.resize((256,256))
        out2 = NG_img.rotate(90)  # 逆时针旋转45度
        out2.save("rotation\\{}\\{}_90_{}.bmp".format(lable[1],lable[1],j))
        NG_img.save("rotation\\{}\\{}_{}.bmp".format(lable[1],lable[1],j))



def load_Img(imgDir):
    lable = os.listdir(imgDir)

    OK_name=os.listdir(imgDir+'/'+lable[0])
    NG_name=os.listdir(imgDir+'/'+lable[1])

    label=[]
    data=[]

    for i in range(len(OK_name)):

        OK_path = imgDir + "/" + lable[0]+'/'+OK_name[i-1]
        OK_img = Image.open(OK_path)
      #  OK_img = OK_img.convert('L')
      #  OK_img = OK_img.resize((64, 64))

        OK_arr = np.asarray(OK_img, dtype="float32")

        OK_arr = OK_arr.flatten()

        data.append(OK_arr)

        label.append(lable[0])

    for j in range(len(NG_name)):

        NG_path = imgDir + '/' + lable[1] + '/' + NG_name[j - 1]
        NG_img = Image.open(NG_path)
       # NG_img = NG_img.convert('L')
       # NG_img = NG_img.resize((64, 64))

        NG_arr = np.asarray(NG_img, dtype="float32")

        NG_arr = NG_arr.flatten()

        data.append(NG_arr)


        label.append(lable[1])
    return label ,data

def main(path,job_name):
   try:
        label, data = load_Img(path)
        X_train, X_test, y_train, y_test = train_test_split(data,label , random_state=33, test_size=0.2)
        pca = PCA(n_components=0.7)

        pca.fit(X_train)

        # 在训练集和测试集降维
        X_train_pca = pca.transform(X_train)
        X_test_pca = pca.transform(X_test)

        SVC4=SVC(C=10000,gamma=1e-06)
        SVC4 = SVC4.fit(X_train_pca, y_train)
        accuracy = SVC4.score(X_test_pca, y_test)
    #保存模型
        name=""+job_name+".pkl"

        pickle.dump((pca,SVC4),open(name, 'wb'))
        return str(accuracy)

   except Exception as err:

       return str(err)


if __name__ == '__main__':
   # print(main("E:\image\SVM"))
    arg1=""+sys.argv[1]
    arg2=""+sys.argv[2]

    gain = Gain_Img(arg1)

    print(main("rotation",arg2))

实时测试

见ＴＣＰ传输图片