CV2识别图片A,B,C,D

这个模型是经过简单训练，只能识别四个英文字母：ABCD，是我为了提取答案才训练的，因为科大，百度，以及腾讯的识别不是很准，所以自己训练自己使用，希望能够给你带来方便。下面是训练好的model,如多你感觉图片的类型相似，可直接使用。

1、图片形式，尽量png

2、模型

如果需要可以加我qq：1851999452

3、

 import os
 import numpy as np
 import cv2
 import matplotlib.pyplot as plt
 import tensorflow as tf
 import keras
 from keras.models import Model,load_model
 from keras.layers import Conv2D,Dense,Activation,Input,MaxPooling2D,Flatten
 from sklearn.model_selection import train_test_split
 # file_dir = '/home/youxue/BorD'
 # B = []
 # A =[]
 # C =[]
 # label_A=[]
 # label_C=[]
 # label_B = []
 # D = []
 # label_D = []
 # for file in os.listdir(file_dir):
 #     if file == 'A':
 #         for img in os.listdir(os.path.join(file_dir, file)):
 #             A.append(file_dir + '/' + file + '/' + img)
 #             label_A.append(0)
 #     elif file == 'B':
 #         for img in os.listdir(os.path.join(file_dir, file)):
 #             B.append(file_dir + '/' + file + '/' + img)
 #             label_B.append(1)
 #     elif file == 'C':
 #         for img in os.listdir(os.path.join(file_dir, file)):
 #             C.append(file_dir + '/' + file + '/' + img)
 #             label_C.append(2)
 #     else:
 #         for img in os.listdir(os.path.join(file_dir, file)):
 #             D.append(file_dir + '/' + file + '/' + img)
 #             label_D.append(3)
 #
 # print('There are %d B\n There are %d D' % (len(B), len(D)))
 #
 # image_list = np.hstack((A,B,C,D))
 # label_list = np.hstack((label_A,label_B,label_C, label_D))
 # temp = np.array([image_list, label_list])
 # temp = temp.transpose()
 #
 #
 # np.random.shuffle(temp)
 # png_list=[]
 # image_list = list(temp[:, 0])
 # label_list = list(temp[:, 1])
 # label_list = [int(i) for i in label_list]
 # for png in image_list:
 #     png_list.append(cv2.imread(png))
 # print(png_list[100].shape)
 # png = np.array(png_list)
 # label = np.array(label_list)
 # x_train,x_test,y_train,y_test=train_test_split(png,label,test_size=0.1)
 #
 # # plt.imshow(x_train[11])
 # # plt.show()
 #
 # x_train = x_train/255.
 # x_test = x_test/255.
 #
 # y_train = np.eye(4)[y_train.reshape(-1)]
 # y_test = np.eye(4)[y_test.reshape(-1)]
 # print(x_train.shape,y_train.shape)

 # input_shape= (24,810,3)
 # input_data = Input(input_shape)
 # layer1 = Conv2D(32,(3,3),padding='same',activation='relu')(input_data)
 # layer2 = MaxPooling2D(pool_size=(2,2))(layer1)
 # layer3 = Conv2D(32,(3,3),padding='same',activation='relu')(layer2)
 # layer4 = MaxPooling2D(pool_size=(2,2))(layer3)
 # layer5 = Conv2D(32,(3,3),padding='same',activation='relu')(layer4)
 # layer6 = Flatten()(layer5)
 # layer7 = Dense(4)(layer6)
 # pred= Activation('softmax')(layer7)
 # model= Model(input=input_data,output=pred)
 # model.summary()
 # model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])
 # model.fit(x_train,y_train,batch_size=32,epochs=20,validation_data=(x_test,y_test))
 # model.save('ABCD.model')
 model=load_model('ABCD.model')
 #这里填写自己的存放路径
 with open('id.txt','a+') as f:
     #这里填写你需要识别图片的路径
     path=''
     for png in os.listdir(path):
         pn= cv2.imread(os.path.join(path,png))
         ids_1 = png.split('answer')[0]
         print(type(ids_1))
     #png2= cv2.imread('/home/youxue/4907988answer.png')
         # plt.imshow(pn)
         # plt.show()
         img=[]
         img.append(pn)
     #img.append(png2)
         img=np.array(img)
         img = img/255.

     # print(img.shape)
         pre=model.predict(img)
         for c in pre:
             temp= max(c[0],c[1],c[2],c[3])
             if temp == c[0]:
                 ABCD = 'A'
             elif temp == c[1]:
                 ABCD = 'B'
             elif temp == c[2]:
                 ABCD = 'C'
             else:
                 ABCD = 'D'

         f.write(ids_1 + '--' + ABCD + '\n')