項目來自唐老師貓狗識別項目及數據集。
項目具體實施步驟:
1.讀取貓狗數據訓練集500+500。
2.對讀取的圖片進行處理,處理成統一大小格式,分好標簽。
3.shuffle一下,將貓狗數據摻雜混合,盡可能隨機。
4.采用CNN網絡訓練測試。
具體代碼如下:
1.讀取訓練集。
import pandas as pd import numpy as np import os import glob import matplotlib.pyplot as plt import cv2 as cv2
images = [] labels = [] img_names = [] cls = [] train_path="training_data" classes = ['dogs','cats'] num_classes = len(classes) image_size=128 print('Going to read training images') for fields in classes: index = classes.index(fields) print('Now going to read {} files (Index: {})'.format(fields, index)) path = os.path.join(train_path, fields, '*g') files = glob.glob(path) print(len(files)) for fl in files: image = cv2.imread(fl) image = cv2.resize(image, (image_size, image_size),0,0, cv2.INTER_LINEAR) image = image.astype(np.float32) image = np.multiply(image, 1.0 / 255.0) images.append(image) label = np.zeros(len(classes)) label[index] = 1.0 labels.append(label) flbase = os.path.basename(fl) img_names.append(flbase) cls.append(fields) images = np.array(images) labels = np.array(labels) img_names = np.array(img_names) cls = np.array(cls)
2.訓練數據集。
from tensorflow import keras from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense,Dropout,Flatten,Conv2D,MaxPool2D from tensorflow.keras import Input from sklearn.model_selection import train_test_split from sklearn.utils import shuffle from tensorflow.keras.regularizers import l1
#數據集處理 images,labels=shuffle(images,labels) X_train, X_test, y_train, y_test=train_test_split(images,labels)
model = Sequential() # 第一個卷積層,32個卷積核,大小5x5,卷積模式SAME,激活函數relu,輸入張量的大小 model.add(Conv2D(filters= 6, kernel_size=(3,3), padding='valid',kernel_regularizer=l1(0.1),activation='tanh',input_shape=(128,128,3))) # model.add(Conv2D(filters= 32, kernel_size=(3,3), padding='valid', activation='relu')) # 池化層,池化核大小2x2 model.add(MaxPool2D(pool_size=(2,2))) # 隨機丟棄四分之一的網絡連接,防止過擬合 model.add(Dropout(0.5)) model.add(Conv2D(filters= 6, kernel_size=(3,3), padding='Same', activation='tanh')) # model.add(Conv2D(filters= 6, kernel_size=(3,3), padding='Same', activation='tanh')) model.add(MaxPool2D(pool_size=(2,2), strides=(2,2))) model.add(Dropout(0.5)) # 全連接層,展開操作, model.add(Flatten()) # 添加隱藏層神經元的數量和激活函數 # model.add(Dense(120, activation='tanh')) model.add(Dropout(0.5)) # model.add(Dense(84, activation='tanh')) # 輸出層 model.add(Dense(2, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(images,labels,validation_split=0.2,batch_size=128,epochs=50)
#模型存儲 mp = "model_3_1625.h5" model.save(mp)
#模型評價 model.evaluate(X_test,y_test)
訓練過程反饋如下:
評價結果:
實驗心得(小白):
1.validation數據集,最好的評價指標達到0.64,隨着epoch訓練的增多,明顯出現過擬合的現象,准確率很多有小於0.5。
2.為了減少過擬合的影響,加入了正則化L1項目與dropout,有一些小的提升。
3.針對optimizer,...............
4.針對batch_size,設置了32,128,100,batch_size設置的小,訓練accuracy上下浮動,不會一直增加
問題:
1.查看第一層次卷積之后的結果
from tensorflow.keras import backend as K layer_1 = K.function([model.layers[0].input], [model.layers[1].output])#第一個 model.layers[0],不修改,表示輸入數據;第二個model.layers[you wanted],修改為你需要輸出的層數的編號 f1 = layer_1([images])[0]#只修改inpu_image # #第一層卷積后的特征圖展示,輸出是(1,149,149,32),(樣本個數,特征圖尺寸長,特征圖尺寸寬,特征圖個數) # for _ in range(32): # show_img = f1[:, :, :, _] # show_img.shape = [63,63,3] # plt.subplot(4, 8, _ + 1) # plt.imshow(show_img, cmap='gray') # plt.axis('off') # plt.show()
model.layers[0].input:128*128*3
model.layers[1].output:63*63*6
a.問什么會是6?
b.查看中間層結果,為什么會出現負值?
結果查看:
plt.figure(figsize=(15, 6.5)) x_axis=range(500) y_axis_data1=history.history["accuracy"] y_axis_data2=history.history["val_accuracy"] plt.plot(x_axis, y_axis_data1, color='red', label="accuracy") plt.plot(x_axis, y_axis_data2, color='blue', label="val_accuracy") plt.hlines(0.65, 0, 500,color="red")#橫線 plt.hlines(0.7, 0, 500,color="red")#橫線 plt.hlines(0.6, 0, 500,color="red")#橫線 plt.show()