CNN圖像識別_算法篇

• 前言
• Keras
• • 1外層循環(huán)
  • 2中部循環(huán)
  • 3內(nèi)部循環(huán)
• Matlab CNN ToolBox
• 總結(jié)

前言

CNN算法方面主要參考的的zh_JNU同學(xué)的工作和Deep-Learning-ToolBox-CNN-master的Matlab源碼，然后也做了些修改和解讀。

Keras

數(shù)據(jù)庫是5鐘分類的400張訓(xùn)練數(shù)據(jù)和100張測試數(shù)據(jù)，數(shù)據(jù)庫網(wǎng)盤（提取碼：f5ze）可能跟環(huán)境版本有關(guān)，我這邊的預(yù)處理不能使用cv的方法，所以統(tǒng)一使用cv2里的方法，值得強調(diào)的是，Keras版本亦有差異，我這邊的版本大致如圖

首先是預(yù)處理，利用opencv對圖片進(jìn)行縮放和灰化，將RGB3通道降維到單通道，python果然是“玩弄”字符串的語言

#coding:utf8 import os import cv2# 遍歷指定目錄，顯示目錄下的所有文件名 #width_scale = 192 #縮放尺寸寬度 #height_scale = 128#縮放尺寸高度 write_path = "/data_base/all_data/test_scale/"#要寫入的圖片路徑"/data_base/all_data/train_scale/"#遍歷每一張圖片進(jìn)行處理 def eachFile(filepath):pathDir = os.listdir(filepath)for allDir in pathDir:child = os.path.join('%s%s' % (filepath,allDir))write_child = os.path.join('%s%s' % (write_path,allDir))image = cv2.imread(child,cv2.IMREAD_GRAYSCALE)#image = cv.LoadImage(child,0)shrink = cv2.resize(image,(126,126),interpolation=cv2.INTER_AREA) #(192,128)cv2.imwrite(write_child,shrink) # break if __name__ == '__main__':filePathC = "/data_base/all_data/test/"##"/data_base/all_data/train/" 訓(xùn)練測試數(shù)據(jù)集分別處理eachFile(filePathC)

這里的預(yù)處理后的維度調(diào)整為126*126，主要考慮的是每層池化層的輸入都為偶數(shù)，卷積核kernel size 為3，那么，輸入層126*126 --> 1卷積層124*124–>1池化層62*62–>2卷積層60*60–>2池化層30*30–>3卷積層28*28–>3池化層14*14，接下來就是全連接，預(yù)處理后

然后是測試，包括網(wǎng)絡(luò)的訓(xùn)練和測試，我這里除了版本兼容上的調(diào)整，也做了些改動，
首先是修改后的源碼，這里有一些術(shù)語，我將它們梳理一下

1外層循環(huán)

epoch：完整樣本訓(xùn)練的次數(shù)；
batch：1個樣本空間中樣本的個數(shù)；
有時1個完整的樣本空間可能包含大量的樣本，這時需要對樣本進(jìn)行分割，分成minibatch，那么完成 一次樣本訓(xùn)練就需要迭代iteration = batch/minibatch；minibatch的一種極限情形就是等于1，理論上不是不可以，但是容易過擬合。

2中部循環(huán)

filters：某一個卷積層中卷積核的個數(shù)，之后再經(jīng)過池化層后，每一個filter都會對應(yīng)一個feature map（特征圖）

3內(nèi)部循環(huán)

kernel_size：卷積核的維數(shù)，在滑動過程中也會有stride、padding等參數(shù)的選取
label:標(biāo)簽的維數(shù)，其實我覺得標(biāo)簽的選取還是挺自由的，可以認(rèn)為每個分類就是一個實數(shù)，也可以是種類維度的獨熱碼向量，如果是1*1的實數(shù)R，在BP過程可能會簡單點，因為均方誤差也是一個實數(shù)，獨熱碼向量的形式，其實已經(jīng)被賦予了一定的意義，即每一位代表了樣本為該類的可能性（probability）

#coding:utf8 import re import cv2 import os import numpy as npfrom keras.models import Sequential from keras.layers.advanced_activations import LeakyReLU from keras.layers.core import Dense,Dropout,Activation,Flatten from keras.layers.convolutional import Conv2D,MaxPooling2D from keras.optimizers import Adam from keras.utils import np_utils#得到一共多少個樣本 def getnum(file_path):pathDir = os.listdir(file_path)i = 0for allDir in pathDir:i +=1return i #制作數(shù)據(jù)集 def data_label(path,count):data = np.empty((count,1,126,126),dtype = 'float32')#建立空的四維張量類型32位浮點label = np.empty((count,),dtype = 'uint8')i = 0pathDir = os.listdir(path)for each_image in pathDir:all_path = os.path.join('%s%s' % (path,each_image))#路徑進(jìn)行連接image = cv2.imread(all_path,0)mul_num = re.findall(r"\d",all_path)#尋找字符串中的數(shù)字，由于圖像命名為300.jpg 標(biāo)簽設(shè)置為0num = int(mul_num[0])-3 # print num,each_image # cv2.imshow("fad",image) # print childarray = np.asarray(image,dtype='float32')array -= np.min(array)array /= np.max(array)data[i,:,:,:] = arraylabel[i] = int(num)i += 1return data,label #構(gòu)建卷積神經(jīng)網(wǎng)絡(luò) def cnn_model(train_data,train_label,test_data,test_label):model = Sequential() #卷積層 12 × 124 × 124 大小model.add(Conv2D(filters = 12,kernel_size = (3,3),padding = "valid",data_format = "channels_first",input_shape = (1,126,126)))model.add(Activation('relu'))#激活函數(shù)使用修正線性單元 #池化層12 × 62 × 62model.add(MaxPooling2D(pool_size = (2,2),strides = (2,2),padding = "valid")) #卷積層 24 * 60 * 60model.add(Conv2D(24,(3,3),padding = "valid",data_format = "channels_first"))model.add(Activation('relu')) #池化層 24×30×30model.add(MaxPooling2D(pool_size = (2,2),strides = (2,2),padding = "valid")) #卷積層 48×28×28model.add(Conv2D(48,(3,3),padding = "valid",data_format = "channels_first"))model.add(Activation('relu')) #池化層 48×14×14model.add(MaxPooling2D(pool_size = (2,2),strides =(2,2),padding = "valid"))model.add(Flatten())model.add(Dense(20))model.add(Activation(LeakyReLU(0.3)))model.add(Dropout(0.5))model.add(Dense(20))model.add(Activation(LeakyReLU(0.3)))model.add(Dropout(0.4))model.add(Dense(5,kernel_initializer="normal"))model.add(Activation('softmax'))adam = Adam(lr = 0.001)model.compile(optimizer = adam,loss = 'categorical_crossentropy',metrics = ['accuracy'])print ('----------------training-----------------------')model.fit(train_data,train_label,batch_size = 20,epochs = 50,shuffle = "True",validation_split = 0.1)print ('----------------testing------------------------')loss,accuracy = model.evaluate(test_data,test_label)print ('\n test loss:',loss)print ('\n test accuracy',accuracy) train_path = "/data_base/all_data/train_scale/" test_path = "/data_base/all_data/test_scale/" train_count = getnum(train_path) test_count = getnum(test_path) train_data,train_label = data_label(train_path,train_count) test_data,test_label = data_label(test_path,test_count) train_label = np_utils.to_categorical(train_label,num_classes = 5) test_label = np_utils.to_categorical(test_label,num_classes = 5) cnn_model(train_data,train_label,test_data,test_label)#print getnum('/home/zhanghao/data/classification/test_scale/') #data_label('/home/zhanghao/data/classification/test_scale/',1) #cv.WaitKey(0)

代碼是3層卷積+池化再進(jìn)行全連接，因為kernel_size是33，并且沒有對輸入樣本擴展，所以卷積層后的維數(shù)就是Nin-kernel_size+1，池化層是22，所以池化后寬高減半。

查看中間變量的方法，插入觀察層，首先要給每一層都取個名字，目前，它的這個打印結(jié)果有點令人撓頭

#coding:utf8 import re import cv2 import os import numpy as npfrom keras.models import Sequential from keras.layers.advanced_activations import LeakyReLU from keras.layers.core import Dense,Dropout,Activation,Flatten from keras.layers.convolutional import Conv2D,MaxPooling2D from keras.optimizers import Adam from keras.utils import np_utils from keras.layers import Densefrom keras.models import Model #得到一共多少個樣本 def getnum(file_path):pathDir = os.listdir(file_path)i = 0for allDir in pathDir:i +=1return i #制作數(shù)據(jù)集 def data_label(path,count):data = np.empty((count,1,126,126),dtype = 'float32')#建立空的四維張量類型32位浮點label = np.empty((count,),dtype = 'uint8')i = 0pathDir = os.listdir(path)for each_image in pathDir:all_path = os.path.join('%s%s' % (path,each_image))#路徑進(jìn)行連接image = cv2.imread(all_path,0)mul_num = re.findall(r"\d",all_path)#尋找字符串中的數(shù)字，由于圖像命名為300.jpg 標(biāo)簽設(shè)置為0num = int(mul_num[0])-3 # print num,each_image # cv2.imshow("fad",image) # print childarray = np.asarray(image,dtype='float32')array -= np.min(array)array /= np.max(array)data[i,:,:,:] = arraylabel[i] = int(num)i += 1return data,label #構(gòu)建卷積神經(jīng)網(wǎng)絡(luò) def cnn_model(train_data,train_label,test_data,test_label):model = Sequential() #卷積層 12 × 120 × 120 大小model.add(Conv2D(filters = 12,kernel_size = (3,3),padding = "valid",data_format = "channels_first",input_shape = (1,126,126),name="cnn1"))model.add(Activation('relu'))#激活函數(shù)使用修正線性單元 #池化層12 × 60 × 60model.add(MaxPooling2D(pool_size = (2,2),strides = (2,2),padding = "valid",name="mxp1")) #卷積層 24 * 58 * 58model.add(Conv2D(24,(3,3),padding = "valid",data_format = "channels_first",name="cnn2"))model.add(Activation('relu')) #池化層 24×29×29model.add(MaxPooling2D(pool_size = (2,2),strides = (2,2),padding = "valid",name="mxp2")) #卷積層 model.add(Conv2D(48,(3,3),padding = "valid",data_format = "channels_first",name="cnn3"))model.add(Activation('relu')) #池化層 model.add(MaxPooling2D(pool_size = (2,2),strides =(2,2),padding = "valid",name="mxp3"))model.add(Flatten())model.add(Dense(20))model.add(Activation(LeakyReLU(0.3)))model.add(Dropout(0.5))model.add(Dense(20))model.add(Activation(LeakyReLU(0.3)))model.add(Dropout(0.4))model.add(Dense(5,kernel_initializer="normal"))model.add(Activation('softmax'))adam = Adam(lr = 0.001)model.compile(optimizer = adam,loss = 'categorical_crossentropy',metrics = ['accuracy'])print ('----------------training-----------------------')model.fit(train_data,train_label,batch_size = 20,epochs = 1,shuffle = "True",validation_split = 0.1)print ('----------------testing------------------------')loss,accuracy = model.evaluate(test_data,test_label)print ('\n test loss:',loss)print ('\n test accuracy',accuracy)##注意縮進(jìn) 觀察層應(yīng)為模型之中dense1_layer_model = Model(inputs=model.input,outputs=model.get_layer("cnn2").output)dense1_output = dense1_layer_model.predict(train_data)print (dense1_output.shape)weight_Dense_1,bias_Dense_1 = model.get_layer("cnn2").get_weights() print(weight_Dense_1.shape)print(bias_Dense_1.shape)print(weight_Dense_1)print(bias_Dense_1)train_path = "/data_base/all_data/train_scale/" test_path = "/data_base/all_data/test_scale/" train_count = getnum(train_path) test_count = getnum(test_path) train_data,train_label = data_label(train_path,train_count) test_data,test_label = data_label(test_path,test_count) train_label = np_utils.to_categorical(train_label,num_classes = 5) test_label = np_utils.to_categorical(test_label,num_classes = 5) cnn_model(train_data,train_label,test_data,test_label)##錯誤 #取某一層的輸出為輸出新建為model，采用函數(shù)模型 #dense1_layer_model = Model(inputs=model.input, # outputs=model.get_layer('cnn1').output) #以這個model的預(yù)測值作為輸出 #dense1_output = dense1_layer_model.predict(data)#print getnum('/home/zhanghao/data/classification/test_scale/') #data_label('/home/zhanghao/data/classification/test_scale/',1) #cv.WaitKey(0)train_path = "/data_base/all_data/train_scale/" test_path = "/data_base/all_data/test_scale/" train_count = getnum(train_path) test_count = getnum(test_path) train_data,train_label = data_label(train_path,train_count) test_data,test_label = data_label(test_path,test_count) train_label = np_utils.to_categorical(train_label,num_classes = 5) test_label = np_utils.to_categorical(test_label,num_classes = 5) cnn_model(train_data,train_label,test_data,test_label)

Matlab CNN ToolBox

這個工具箱最早是在Github上開源的，論壇上也有很多同學(xué)對源碼進(jìn)行解讀，我覺得都挺好的，不過，我覺得最好的輔助理解是Sunshine同學(xué)寫的兩層CNN的Matlab實現(xiàn)，他把前一個卷積+池化和后一個卷積+池化之間的前饋和反饋連接關(guān)系講得很清楚，并且，和這個CNN ToolBox的處理方式一致，讀者能夠很直觀地看到各個子層的維數(shù)變化

總結(jié)

看了挺多資料的，覺得把這兩個例子讀通，會很有幫助，尤其是Matlab的這個，你可以看到很多細(xì)節(jié)的處理方式

總結(jié)

以上是生活随笔為你收集整理的CNN图像识别_算法篇的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。