# -*- coding: utf-8 -*- """ Spyder Editor This is a temporary script file. python 3.6 pip install tensorflow-gpu==2.6.0 pip install keras==2.6.0 pip install opencv-python==3.4.2.16 pip install matplotlib==3.3.4 """ import numpy as np import os # import cv2 from sklearn.model_selection import train_test_split # pip install scikit-learn from tensorflow import keras from tensorflow.keras import layers from tensorflow.keras import models from tensorflow.keras.preprocessing.image import img_to_array from tensorflow.keras.preprocessing.image import load_img import tensorflow as tf import os class LearningRateTracker(keras.callbacks.Callback): def __init__(self): self._lr = list() def on_epoch_end(self, epoch, logs=None): optimizer = self.model.optimizer self._lr.append(tf.keras.backend.get_value(optimizer.lr)) print(f'Epoch {epoch + 1}: Learning rate is {tf.keras.backend.get_value(optimizer.lr)}.\n') def show_train_learningRate(self): plt.plot(self._lr) plt.title('learningRate') plt.ylabel('learningRate') plt.xlabel('Epoch') plt.legend(['train', 'validation'], loc='upper left') plt.savefig('LearningRate.png') plt.show() # os.environ[''] # ============================================================================= from keras.utils import np_utils # ============================================================================= config = tf.compat.v1.ConfigProto() gpus = tf.config.experimental.list_physical_devices('GPU') os.environ["CUDA_VISIBLE_DEVICES"] = '0' session = tf.compat.v1.Session(config=config) epochs = 150 # 訓練的次數 batch_size = 18 img_rows = None # 驗證碼影像檔的高 img_cols = None # 驗證碼影像檔的寬 # digits_in_img = 6 #驗證碼影像檔中有幾位數 x_list = list() # 存所有驗證碼數字影像檔的array y_list = list() # 存所有的驗證碼數字影像檔array代表的正確數字 x_train = list() # 存訓練用驗證碼數字影像檔的array y_train = list() # 存訓練用驗證碼數字影像檔array代表的正確數字 x_test = list() # 存測試用驗證碼數字影像檔的array y_test = list() # 存測試用驗證碼數字影像檔array代表的正確數字 def split_digits_in_img(img_array, x_list, y_list=None): # 副函式：分割及儲存驗證碼 # for i in range(digits_in_img): # step = img_cols // digits_in_img #step=圖片總寬度除六 # y_list.append(img_filename[:2]) #將圖片正確數字(檔名)存進 y_list x_list.append(img_array) # 將圖片存進x_list input_filepath = './trainingset/1205_Trainning' file_list = sorted(os.listdir(input_filepath)) # 获取文件名列表. print(f'file_list = {file_list}') for classnumber in file_list: img_filenames = os.listdir(f'{input_filepath}/{classnumber}') count = 0 # os.listdir() 方法用于返回指定的文件夹包含的文件或文件夹的名字的列表 for img_filename in img_filenames: count += 1 if '.png' not in img_filename: # 只讀取PNG檔 continue # img = load_img('./output_file(00)/{0}/{1}'.format(classnumber, img_filename), color_mode='rgb', target_size=(224, 224, 3)) #將圖片依序以灰階讀取 img = load_img('{0}/{1}/{2}'.format(input_filepath, classnumber, img_filename), color_mode='grayscale', target_size=(112, 112, 1)) # 將圖片依序以灰階讀取 img_array = img_to_array(img) # 將圖片轉換成數值 img_rows, img_cols, _ = img_array.shape # 將圖片長寬分別儲存在img_rows, img_cols, # print(img_array.shape) y_list.append(classnumber[:2]) # 將圖片正確數字(檔名)存進 y_list split_digits_in_img(img_array, x_list, y_list) # 呼叫 副函式：分割及儲存驗證碼 print(f'{classnumber} is ok --> 有{count}張') x_train, x_test, y_train, y_test = train_test_split(x_list, y_list, test_size=0.1, stratify=y_list) # , stratify=y_list随机划分训练集和测试集 # print(x_train) # print(y_train) y_train = keras.utils.to_categorical(y_train) # 將y_list 轉換成categorical形式 y_test = keras.utils.to_categorical(y_test) # ============================================================================= # y_train_onehot = np_utils.to_categorical(y_train) # y_test_onehot = np_utils.to_categorical(y_test) # ============================================================================= # ============================================================================= # if os.path.isfile('cnn_model.h5'): #檢測是否已有模型 # model = models.load_model('cnn_model.h5') #將該模型導入 # print('Model loaded from file.') # else: #否則：建立新模型 # ============================================================================= model = models.Sequential() # 創建模型 model.add(layers.Conv2D(input_shape=(112, 112, 1), filters=64, kernel_size=(4, 4), padding="same", activation="relu")) model.add(layers.Conv2D(filters=64, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.MaxPool2D(pool_size=(3, 3), strides=(2, 2))) model.add(layers.Dropout(rate=0.3)) # 0 model.add(layers.Conv2D(filters=128, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=128, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.MaxPool2D(pool_size=(3, 3), strides=(2, 2))) model.add(layers.Dropout(rate=0.2)) # 0 model.add(layers.Conv2D(filters=256, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=256, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=256, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=256, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.MaxPool2D(pool_size=(3, 3), strides=(2, 2))) model.add(layers.Dropout(rate=0.2)) # 0 model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.MaxPool2D(pool_size=(3, 3), strides=(2, 2))) model.add(layers.Dropout(rate=0.2)) # 5 model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.Conv2D(filters=512, kernel_size=(3, 3), padding="same", activation="relu")) model.add(layers.MaxPool2D(pool_size=(3, 3), strides=(2, 2))) model.add(layers.Flatten()) model.add(layers.Dense(units=4096, activation="relu")) model.add(layers.Dropout(rate=0.5)) # 6 model.add(layers.Dense(units=4096, activation="relu")) model.add(layers.Dense(units=17, activation="softmax")) # 這是類別數 # 全連接層 (輸出空間的維數,activation) print('New model created.') # 全連接層 (輸出空間的維數,activation) model.compile(loss=keras.losses.categorical_crossentropy, optimizer=keras.optimizers.Adam(learning_rate=0.000001, beta_1=0.9, beta_2=0.999), metrics=['accuracy']) # model.compile 編譯模型（optimizer =優化器，loss =損失函數， metrics= [“準確率”]） # model.fit(np.array(x_train), np.array(y_train), batch_size=digits_in_img, epochs=epochs, verbose=1, validation_data=(np.array(x_test), np.array(y_test))) # ============================================================================= checkpoint_filepath = os.path.sep.join(['checkpoint', "weights-{epoch:03d}-{loss:.9f}-{val_accuracy:.4f}.h5"]) model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint( filepath=checkpoint_filepath, save_weights_only=False, monitor='loss', mode='min', save_best_only=True) lr_tracker = LearningRateTracker() with tf.device('/cpu:0'): x_train = tf.convert_to_tensor(x_train, np.float32) y_train = tf.convert_to_tensor(y_train, np.float32) train_history = model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1, callbacks=[model_checkpoint_callback, lr_tracker], validation_data=(np.array(x_test), np.array(y_test)), shuffle=True) import matplotlib.pyplot as plt def show_train_history(train_history, title, train_metric, val_metric): plt.plot(train_history.history[train_metric]) plt.plot(train_history.history[val_metric]) plt.title(title) plt.ylabel(train_metric) plt.xlabel('Epoch') plt.legend(['train', 'validation'], loc='upper left') plt.savefig(f'{title}.png') plt.show() show_train_history(train_history, 'Accuracy', 'accuracy', 'val_accuracy') show_train_history(train_history, 'Loss', 'loss', 'val_loss') lr_tracker.show_train_learningRate() # ============================================================================= loss, accuracy = model.evaluate(np.array(x_test), np.array(y_test), verbose=1) # 評估模型(verbose：0或1。詳細模式。 0 =靜音，1 =進度欄。) # print('Test loss:', loss) # print('Test accuracy:', accuracy) model.save('model1124.h5') print(model.summary())