Python機器學習設計——果蔬識別

一、選題背景

在學習了深度學習案例——MNIST手寫數字識別和基於keras框架的貓狗圖像識別，為了進一步熟悉tensorflow和keras的基本用法和網絡框架，試想着實現對水果和蔬菜的種類進行識別。

二、機器學習案例設計方案

1.數據集來源

水果蔬菜數據集:

數據集來源於kaggle，因數據集作者為了構建一個應用程序，該應用程序可以從捕獲的照片中識別食品，並為用戶提供可以使用食品制作的不同食譜從必應圖像搜索中抓取的，共包含4291張圖像，36個種類，分為train(每個種類100張)，test(每個種類10張)，validation(每個種類10張)三個文件夾，上述每個文件夾都包含不同水果和蔬菜的子文件夾，其中存在相應食品的圖像，包含水果10種，蔬菜26種，共36種。

2.采用的機器學習框架描述

卷積神經網絡(Convolutional Neural Network, CNN)：卷積神經網絡是一類包含卷積計算且具有深度結構的前饋神經網絡，是深度學習技術中極具代表的網絡結構之一，在圖像處理領域取得了很大的成功，在國際標准的 ImageNet 數據集上，許多成功的模型都是基於 CNN 的。

Tensorflow：TensorFlow是谷歌基於DistBelief進行研發的第二代人工智能學習系統，其命名來源於本身的運行原理。Tensor（張量）意味着N維數組，Flow（流）意味着基於數據流圖的計算，TensorFlow為張量從流圖的一端流動到另一端計算過程。TensorFlow是將復雜的數據結構傳輸至人工智能神經網中進行分析和處理過程的系統，可被用於語音識別或圖像識別等多項機器學習和深度學習領域。

Keras：Keras是一個模型級( model-level)的庫，為開發深度學習模型提供了高層次的構建模塊。它不處理張量操作、求微積分等基礎的運算，而是依賴--個專門的、高度優化的張量庫來完成這些運算。這個張量庫就是Keras的后端引擎(backend engine)，如TensorFlow等。

3.涉及到的技術難點和解決思路

一些第三方庫沒有此模塊的問題  解決思路：降版本或者使用功能相同的新模塊

三、機器學習的實現步驟

果蔬識別

1.對原有數據集進行分類

1）將數據集train，test，和validation分類成fruit和vegetable兩類

import os

import shutil

#訓練集數據處理
def train(file_path):
    d=[]
    s=[]
    for root, dirs , files in os.walk(file_path): #讀取文件並提取出文件路徑和類名    
        d.append(root)#文件路徑
        s.append(dirs)#類名
        for i in s:
            if i!=s[0]:
                s.remove(i)
        b = [ i for item in s for i in item]        
    d.pop(0)
    fruit=[ 'banana', 'apple', 'pear', 'grapes', 'orange','kiwi','watermelon', 'pomegranate', 'pineapple', 'mango']
    vegetables=['cucumber', 'carrot', 'capsicum', 'onion', 'potato', 'lemon', 'tomato', 
                'raddish', 'beetroot', 'cabbage', 'lettuce', 'spinach', 'soy beans', 
                'cauliflower', 'bell pepper', 'chilli pepper', 'turnip', 'corn', 'sweetcorn',
                'sweetpotato', 'paprika', 'jalepeno', 'ginger', 'garlic', 'peas', 'eggplant']
    #用for循環對水果和蔬菜進行分類
    #訓練集數據處理
    #水果
    for i in b:
        for j in fruit:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/train",i),
                            "C:/Users/linyicheng/Desktop/fruit/train")
    #蔬菜
    for i in b:
        for j in vegetables:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/train",i),
                            "C:/Users/linyicheng/Desktop/vegetable/train")

train("C:/Users/linyicheng/Desktop/train")

#驗證集數據處理
def validation(file_path):
    d=[]
    s=[]
    for root, dirs , files in os.walk(file_path):     
        d.append(root)
        s.append(dirs)
        for i in s:
            if i!=s[0]:
                s.remove(i)
        b = [ i for item in s for i in item]        
    d.pop(0)
    fruit=[ 'banana', 'apple', 'pear', 'grapes', 'orange','kiwi','watermelon', 'pomegranate', 'pineapple', 'mango']
    vegetables=['cucumber', 'carrot', 'capsicum', 'onion', 'potato', 'lemon', 'tomato', 
                'raddish', 'beetroot', 'cabbage', 'lettuce', 'spinach', 'soy beans', 
                'cauliflower', 'bell pepper', 'chilli pepper', 'turnip', 'corn', 'sweetcorn',
                'sweetpotato', 'paprika', 'jalepeno', 'ginger', 'garlic', 'peas', 'eggplant']
    #訓練集數據處理
    #水果
    for i in b:
        for j in fruit:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/validation",i),
                            "C:/Users/linyicheng/Desktop/fruit/validation")
    #蔬菜
    for i in b:
        for j in vegetables:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/validation",i),
                            "C:/Users/linyicheng/Desktop/vegetable/validation")

validation("C:/Users/linyicheng/Desktop/validation")

#測試集數據處理
def test(file_path):
    d=[]
    s=[]
    for root, dirs , files in os.walk(file_path):     
        d.append(root)
        s.append(dirs)
        for i in s:
            if i!=s[0]:
                s.remove(i)
        b = [ i for item in s for i in item]        
    d.pop(0)
    fruit=[ 'banana', 'apple', 'pear', 'grapes', 'orange','kiwi','watermelon', 'pomegranate', 'pineapple', 'mango']
    vegetables=['cucumber', 'carrot', 'capsicum', 'onion', 'potato', 'lemon', 'tomato', 
                'raddish', 'beetroot', 'cabbage', 'lettuce', 'spinach', 'soy beans', 
                'cauliflower', 'bell pepper', 'chilli pepper', 'turnip', 'corn', 'sweetcorn',
                'sweetpotato', 'paprika', 'jalepeno', 'ginger', 'garlic', 'peas', 'eggplant']
    #訓練集數據處理
    #水果
    for i in b:
        for j in fruit:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/test",i),
                            "C:/Users/linyicheng/Desktop/fruit/test")
    #蔬菜
    for i in b:
        for j in vegetables:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/test",i),
                            "C:/Users/linyicheng/Desktop/vegetable/test")

test("C:/Users/linyicheng/Desktop/test")

2.對數據集進行預處理，模仿貓狗識別建立數據集目錄

1）對fruit，vegetable里的train，test，validation進行重命名

#水果
outer_path ="C:/Users/linyicheng/Desktop/fruit/test"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)        
                i += 1
            except:
                continue

outer_path ="C:/Users/linyicheng/Desktop/fruit/train"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue

#驗證集
outer_path ="C:/Users/linyicheng/Desktop/fruit/validation"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue

#蔬菜
outer_path ="C:/Users/linyicheng/Desktop/vegetable/test"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)        
                i += 1
            except:
                continue

outer_path ="C:/Users/linyicheng/Desktop/vegetable/train"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue

outer_path ="C:/Users/linyicheng/Desktop/vegetable/validation"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue

2）將文件移到base目錄中，構建數據集目錄結構

#目標文件夾，此處為相對路徑，也可以改為絕對路徑
import shutil
determination = "C:/Users/linyicheng/Desktop/base/test/fruit"#目標
if not os.path.exists(determination):
    os.makedirs(determination)

#源文件夾路徑
path = "C:/Users/linyicheng/Desktop/fruit/test"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

determination = "C:/Users/linyicheng/Desktop/base/train/fruit"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/fruit/train"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

determination = "C:/Users/linyicheng/Desktop/base/validation/fruit"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/fruit/validation"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

import shutil
determination = "C:/Users/linyicheng/Desktop/base/test/vegetable"#目標
if not os.path.exists(determination):
    os.makedirs(determination)

#源文件夾路徑
path = "C:/Users/linyicheng/Desktop/vegetable/test"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

determination = "C:/Users/linyicheng/Desktop/base/train/vegetable"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/vegetable/train"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

determination = "C:/Users/linyicheng/Desktop/base/validation/vegetable"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/vegetable/validation"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

3.設置路徑

base_dir="C:/Users/linyicheng/Desktop/base/"
train_data_dir="C:/Users/linyicheng/Desktop/base/train/"
test_data_dir="C:/Users/linyicheng/Desktop/base/test/"
val_data_dir="C:/Users/linyicheng/Desktop/base/validation/"
train_fruit_data="C:/Users/linyicheng/Desktop/base/train/fruit/"
test_fruit_data="C:/Users/linyicheng/Desktop/base/test/fruit/"
test_vegetable_data="C:/Users/linyicheng/Desktop/base/test/vegetable/"

4.搭建卷積神經網絡

#搭建卷積神經網絡
from keras import layers
from keras import models
model=models.Sequential()
"""
Output shape計算公式：(輸入尺寸-卷積核尺寸/步長+1
對CNN模型，Param的計算方法如下：
卷積核長度*卷積核寬度*通道數+1）*卷積核個數
輸出圖片尺寸：150-3+1=148*148
參數數量：32*3*3*3+32=896
"""
model.add(layers.Conv2D(32,(3,3),activation='relu',input_shape=(150,150,3)))#卷積層1
model.add(layers.MaxPooling2D(2,2))#最大池化層1
# 輸出圖片尺寸：148/2=74*74 
# 輸出圖片尺寸：74-3+1=72*72，參數數量：64*3*3*32+64=18496
#32是第1個卷積層的輸出的通道數
model.add(layers.Conv2D(64, (3, 3), activation='relu'))#卷積層2
model.add(layers.MaxPooling2D((2, 2)))#最大池化層2
# 輸出圖片尺寸：72/2=36*36
 
#Output Shape的輸出為36
# 輸出圖片尺寸：36-3+1=34*34，參數數量：128*3*3*64+128=73856
model.add(layers.Conv2D(128, (3, 3), activation='relu'))#卷積層3
model.add(layers.MaxPooling2D((2, 2)))#最大池化層3
# 輸出圖片尺寸：34/2=17*17
# 輸出圖片尺寸：17-3+1=15*15，參數數量：128*3*3*128+128=147584
model.add(layers.Conv2D(128, (3, 3), activation='relu'))# 輸出圖片尺寸：15/2=7*7
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))   #全連接層1
model.add(layers.Dense(1, activation='sigmoid'))#全連接層2，作為輸出層

 從輸出看出，卷積神經網絡dense_1 (Dense)的參數總數達到300多萬

model.summary() 

 5.編譯模型

# 編譯模型
# RMSprop 優化器。因為網絡最后一層是單一sigmoid單元
from keras import optimizers
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
               metrics=['acc'])

6.使用 ImageDataGenerator 從目錄中讀取樣本數據

from keras.preprocessing.image import ImageDataGenerator
#歸一化

train_datagen = ImageDataGenerator(rescale=1./255)
validation_datagen=ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)
# 數據集加載函數，指明數據集的位置並統一處理為imgheight*imgwidth的大小，同時設置batch
    # 加載訓練集
train_ds = train_datagen.flow_from_directory (
        #文件目錄位置
        train_data_dir,
#輸入訓練圖像尺寸,所有圖片的size必須是150x150
        class_mode='binary',
        target_size=(150, 150),
        batch_size=20)
    #加載測試集
test_ds = test_datagen.flow_from_directory(
        test_data_dir,
        class_mode='binary',
        target_size=(150, 150),
        batch_size=20)
    # 加載驗證集
val_ds =test_datagen.flow_from_directory(
        val_data_dir,
        class_mode='binary',
        target_size=(150, 150),
        batch_size=20)

for data_batch, labels_batch in train_ds:
    print('data batch shape:', data_batch.shape)
    print('labels batch shape:', labels_batch.shape)
    break

 

 7.訓練模型（30次）

#訓練模型30輪次，可以修改epochs的值
history = model.fit_generator(
       train_ds,
       steps_per_epoch=100,
       epochs=30,
       validation_data=val_ds,
       validation_steps=50)

可以看出訓練集精度（acc）：0.9130 和驗證集精度（val_acc）：0.9803

模型較為成功，接下來通過數據增強來提升精度

 #將訓練過程產生的數據保存為h5文件

model.save('fruit_and_vegetable_30epoch.h5') 

8.繪制訓練過程中的損失曲線和精度曲線

import matplotlib.pyplot as plt
 
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(1, len(acc) + 1)

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure() 
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
 
plt.show()

由圖可知訓練精度隨着時間線性增加，直到接近100%

9.單張圖片進行判斷圖片是水果還是蔬菜

from PIL import Image
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
#加載模型
model=load_model('fruit_and_vegetable_30epoch.h5')
#本地圖片路徑
img_path=("C:/Users/linyicheng/Desktop/base/train/fruit/apple_41.jpg")
img = image.load_img(img_path,target_size=(150,150))
# 將其轉換為具有形狀的Numpy數組（150、150、3）
img_tensor = image.img_to_array(img)/255
# 將其形態變為(150,150,150,3)的形狀
img_tensor = np.expand_dims(img_tensor, axis=0)
#取圖片信息
prediction =model.predict(img_tensor)
#輸出識別率
print(prediction)
if prediction>0.5:
    print('水果')
else:
    print('蔬菜')
plt.imshow(img)

10. 數據增強訓練，利用ImageDataGenerator實現數據增強

from keras.preprocessing.image import ImageDataGenerator
from keras.models import load_model
from tensorflow.python.keras.preprocessing import image
from tensorflow.python.keras.preprocessing.image import ImageDataGenerator,array_to_img,img_to_array,load_img

import os
from keras.preprocessing.image import load_img
import matplotlib.pyplot as plt
figure,ax = plt.subplots(nrows=1, ncols=4, sharex=True, 
                         sharey=True,figsize=(16,10))

datagen=ImageDataGenerator(
    rotation_range=40,      #圖像隨機旋轉角度范圍
    width_shift_range=0.2,  #圖片在水平方向上平移的比例
    height_shift_range=0.2, #圖片在垂直方向上平移的比例
    shear_range=0.2,        #隨機錯切變換的角度
    zoom_range=0.2,         #圖像隨機縮放的范圍
    horizontal_flip=True,   #隨機將一半圖像水平翻轉
      
    fill_mode='nearest')
#填充創建像素的一種方法
fnames=[os.path.join(train_fruit_data,fname)for fname in os.listdir(train_fruit_data)]
img_path=fnames[6]
img=image.load_img(img_path,target_size=(150,150))
x = image.img_to_array(img)/255
x = x.reshape((1,) + x.shape)
i = 0
for batch in datagen.flow(x, batch_size=1):
    ax = ax.flatten()   # 將子圖從多維變成一維
    plt.figure(i)
    imgplot = ax[i-1].imshow(image.array_to_img(batch[0]), cmap='Greys', interpolation='nearest')
    i += 1
    if i % 4 == 0:
        break
plt.show() #繪制增強后的圖像
'''
雖然使用了數據增強技術，但是從輸出的圖像來看，圖片之間還是有很大的相似度，
因為它們均來自同一張原始圖片，並沒有提供新的信息。
為了盡可能消除過擬合，可以在模型中增加一個DDropout層，添加到密集連接分類
器前
'''

11. 在緊密連接的分類器之前為模型添加一個Dropout層

#在緊密連接的分類器之前為模型添加一個Dropout層
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

12.編譯模型

from keras import optimizers
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
               metrics=['acc'])

13.使用數據增強和dropout來訓練我們的網絡

#使用數據增強和dropout來訓練我們的網絡
#歸一化
train_datagen = ImageDataGenerator(
     rescale=1./255,
     rotation_range=40,
     width_shift_range=0.2,
     height_shift_range=0.2,
     shear_range=0.2,
     zoom_range=0.2,
     horizontal_flip=True)
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
     train_data_dir,
     #輸入訓練圖像尺寸,所有圖片的size必須是150x150
     target_size=(150, 150),
     batch_size=32,
     #因為我們使用二元交叉熵損失，我們需要二元標簽
     class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
     val_data_dir,
     target_size=(150, 150),
     batch_size=32,
     class_mode='binary')

 14.訓練模型（100次）

history = model.fit_generator(
       train_generator,
       steps_per_epoch=100,
       epochs=100,
       validation_data=validation_generator,
       validation_steps=50)

 model.save('fruit_and_vegetable_100epoch.h5')  

15.繪制訓練過程中的損失曲線和精度曲線

import matplotlib.pyplot as plt
 
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(1, len(acc) + 1)

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure() 
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
 
plt.show()

 16.使用新的訓練模型進行識別

from keras.preprocessing import image

from keras.models import load_model
import numpy as np
#加載模型
model=load_model('fruit_and_vegetable_100epoch.h5')
#本地圖片路徑
img_path=("C:/Users/linyicheng/Desktop/base/train/fruit/apple_42.jpg")
img = image.load_img(img_path,target_size=(150,150))
# 將其轉換為具有形狀的Numpy數組（150、150、3）
img_tensor = image.img_to_array(img)/255
# 將其形態變為(150,150,150,3)的形狀
img_tensor = np.expand_dims(img_tensor, axis=0)
#取圖片信息
prediction =model.predict(img_tensor)
#輸出識別率
print(prediction)
if prediction>0.5:
    print('水果')
else:
    print('蔬菜')
plt.imshow(img)

17.多張圖片進行識別判斷

 

from PIL import Image
from keras.preprocessing import image
import matplotlib.pyplot as plt
from keras.models import load_model
plt.rcParams['font.family'] = ['SimHei']
figure,ax = plt.subplots(nrows=2, ncols=4, sharex=True, sharey=True,figsize=(24,10))
def convertjpg(jpgfile):
    #將圖片縮小到（150，150）的大小
    img=Image.open(jpgfile)
    try:
        new_img=img.resize((150,150),Image.BILINEAR)
        return new_img
    except Exception as e:
        print(e)
snames=[os.path.join(test_vegetable_data,sname)for sname in os.listdir(test_vegetable_data)]
j=0
for i in snames[0:8] :
    j+=1
    img_show=convertjpg(i)
    img_scale = image.img_to_array(img_show)
    img_scale = img_scale.reshape(1,150,150,3) # 將形狀轉化為(1,150,150,3)
    img_scale = img_scale.astype('float32')/255
    result = model.predict(img_scale)# 預測函數-
    if result>0.5:
        title = '蔬菜'
        label = '正確識別率為：'+str(result)
    else:
        title = '水果'
        label = '正確識別率為：'+str(1-result)
         
    ax = ax.flatten()   # 將子圖從多維變成一維
    ax[j-1].imshow(img_show, cmap='Greys', interpolation='nearest')
     # 子圖標題
    ax[j-1].set_title(title,fontsize=24)
     # 子圖X軸標簽
    ax[j-1].set_xlabel(label,fontsize=24)
     # 去掉刻度
    ax[0].set_xticks([])
    ax[0].set_yticks([]) 
plt.show()

figure,ax = plt.subplots(nrows=2, ncols=4, sharex=True, sharey=True,figsize=(24,10))

fnames=[os.path.join(test_fruit_data,fname)for fname in os.listdir(test_fruit_data)]
j=0
for i in fnames[0:8]: 
    j+=1
    img_show=convertjpg(i)
    img_scale = image.img_to_array(img_show)
    img_scale = img_scale.reshape(1,150,150,3) # 將形狀轉化為(1,150,150,3)
    img_scale = img_scale.astype('float32')/255
    result = model.predict(img_scale)# 預測函數-
    if result>0.5:
        title = '蔬菜'
        label = '正確識別率為：'+ str(result)
    else:
        title = '水果'
        label = '正確識別率為：'+str(1-result)         
    ax = ax.flatten()   # 將子圖從多維變成一維
    ax[j-1].imshow(img_show, cmap='Greys', interpolation='nearest')
     # 子圖標題
    ax[j-1].set_title(title,fontsize=24)
     # 子圖X軸標簽
    ax[j-1].set_xlabel(label,fontsize=24)
    # 去掉刻度
    ax[0].set_xticks([])
    ax[0].set_yticks([])

 全代碼附上

import os

import shutil

#訓練集數據處理
def train(file_path):
    d=[]
    s=[]
    for root, dirs , files in os.walk(file_path): #讀取文件並提取出文件路徑和類名    
        d.append(root)#文件路徑
        s.append(dirs)#類名
        for i in s:
            if i!=s[0]:
                s.remove(i)
        b = [ i for item in s for i in item]        
    d.pop(0)
    fruit=[ 'banana', 'apple', 'pear', 'grapes', 'orange','kiwi','watermelon', 'pomegranate', 'pineapple', 'mango']
    vegetables=['cucumber', 'carrot', 'capsicum', 'onion', 'potato', 'lemon', 'tomato', 
                'raddish', 'beetroot', 'cabbage', 'lettuce', 'spinach', 'soy beans', 
                'cauliflower', 'bell pepper', 'chilli pepper', 'turnip', 'corn', 'sweetcorn',
                'sweetpotato', 'paprika', 'jalepeno', 'ginger', 'garlic', 'peas', 'eggplant']
    #用for循環對水果和蔬菜進行分類
    #訓練集數據處理
    #水果
    for i in b:
        for j in fruit:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/train",i),
                            "C:/Users/linyicheng/Desktop/fruit/train")
    #蔬菜
    for i in b:
        for j in vegetables:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/train",i),
                            "C:/Users/linyicheng/Desktop/vegetable/train")

train("C:/Users/linyicheng/Desktop/train")
#驗證集數據處理
def validation(file_path):
    d=[]
    s=[]
    for root, dirs , files in os.walk(file_path):     
        d.append(root)
        s.append(dirs)
        for i in s:
            if i!=s[0]:
                s.remove(i)
        b = [ i for item in s for i in item]        
    d.pop(0)
    fruit=[ 'banana', 'apple', 'pear', 'grapes', 'orange','kiwi','watermelon', 'pomegranate', 'pineapple', 'mango']
    vegetables=['cucumber', 'carrot', 'capsicum', 'onion', 'potato', 'lemon', 'tomato', 
                'raddish', 'beetroot', 'cabbage', 'lettuce', 'spinach', 'soy beans', 
                'cauliflower', 'bell pepper', 'chilli pepper', 'turnip', 'corn', 'sweetcorn',
                'sweetpotato', 'paprika', 'jalepeno', 'ginger', 'garlic', 'peas', 'eggplant']
    #訓練集數據處理
    #水果
    for i in b:
        for j in fruit:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/validation",i),
                            "C:/Users/linyicheng/Desktop/fruit/validation")
    #蔬菜
    for i in b:
        for j in vegetables:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/validation",i),
                            "C:/Users/linyicheng/Desktop/vegetable/validation")

validation("C:/Users/linyicheng/Desktop/validation")
#測試集數據處理
def test(file_path):
    d=[]
    s=[]
    for root, dirs , files in os.walk(file_path):     
        d.append(root)
        s.append(dirs)
        for i in s:
            if i!=s[0]:
                s.remove(i)
        b = [ i for item in s for i in item]        
    d.pop(0)
    fruit=[ 'banana', 'apple', 'pear', 'grapes', 'orange','kiwi','watermelon', 'pomegranate', 'pineapple', 'mango']
    vegetables=['cucumber', 'carrot', 'capsicum', 'onion', 'potato', 'lemon', 'tomato', 
                'raddish', 'beetroot', 'cabbage', 'lettuce', 'spinach', 'soy beans', 
                'cauliflower', 'bell pepper', 'chilli pepper', 'turnip', 'corn', 'sweetcorn',
                'sweetpotato', 'paprika', 'jalepeno', 'ginger', 'garlic', 'peas', 'eggplant']
    #訓練集數據處理
    #水果
    for i in b:
        for j in fruit:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/test",i),
                            "C:/Users/linyicheng/Desktop/fruit/test")
    #蔬菜
    for i in b:
        for j in vegetables:
            if i==j:
                shutil.move(os.path.join("C:/Users/linyicheng/Desktop/test",i),
                            "C:/Users/linyicheng/Desktop/vegetable/test")

test("C:/Users/linyicheng/Desktop/test")
#水果
outer_path ="C:/Users/linyicheng/Desktop/fruit/test"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)        
                i += 1
            except:
                continue
outer_path ="C:/Users/linyicheng/Desktop/fruit/train"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue
outer_path ="C:/Users/linyicheng/Desktop/fruit/validation"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue
#蔬菜
outer_path ="C:/Users/linyicheng/Desktop/vegetable/test"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)        
                i += 1
            except:
                continue
outer_path ="C:/Users/linyicheng/Desktop/vegetable/train"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue
outer_path ="C:/Users/linyicheng/Desktop/vegetable/validation"
folderlist = os.listdir(outer_path)          #列舉文件夾
 
for folder in folderlist:     
    inner_path = os.path.join(outer_path, folder)
    total_num_folder = len(folderlist)       #文件夾的總數
       #打印文件夾的總數
 
    filelist = os.listdir(inner_path)        #列舉圖片
    i = 0
    for item in filelist:
        total_num_file = len(filelist)       #單個文件夾內圖片的總數
        if item.endswith('.jpg'):
            src = os.path.join(os.path.abspath(inner_path), item)           #原圖的地址
            dst = os.path.join(os.path.abspath(inner_path), str(folder) + '_' + str(i) + '.jpg')        #新圖的地址（這里可以把str(folder) + '_' + str(i) + '.jpg'改成你想改的名稱）
            try:
                os.rename(src, dst)
                
                i += 1
            except:
                continue
#目標文件夾，此處為相對路徑，也可以改為絕對路徑
import shutil
determination = "C:/Users/linyicheng/Desktop/base/test/fruit"#目標
if not os.path.exists(determination):
    os.makedirs(determination)

#源文件夾路徑
path = "C:/Users/linyicheng/Desktop/fruit/test"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)
determination = "C:/Users/linyicheng/Desktop/base/train/fruit"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/fruit/train"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)
determination = "C:/Users/linyicheng/Desktop/base/validation/fruit"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/fruit/validation"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)
import shutil
determination = "C:/Users/linyicheng/Desktop/base/test/vegetable"#目標
if not os.path.exists(determination):
    os.makedirs(determination)

#源文件夾路徑
path = "C:/Users/linyicheng/Desktop/vegetable/test"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)
determination = "C:/Users/linyicheng/Desktop/base/train/vegetable"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/vegetable/train"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)
determination = "C:/Users/linyicheng/Desktop/base/validation/vegetable"
if not os.path.exists(determination):
    os.makedirs(determination)
path = "C:/Users/linyicheng/Desktop/vegetable/validation"
folders = os.listdir(path)
for folder in folders:
    dir = path + '/' + str(folder)
    files = os.listdir(dir)
    for file in files:
        source = dir + '/' + str(file)
        deter = determination + '/' + str(file)
        shutil.copyfile(source, deter)

base_dir="C:/Users/linyicheng/Desktop/base/"
train_data_dir="C:/Users/linyicheng/Desktop/base/train/"
test_data_dir="C:/Users/linyicheng/Desktop/base/test/"
val_data_dir="C:/Users/linyicheng/Desktop/base/validation/"
train_fruit_data="C:/Users/linyicheng/Desktop/base/train/fruit/"
test_fruit_data="C:/Users/linyicheng/Desktop/base/test/fruit/"
test_vegetable_data="C:/Users/linyicheng/Desktop/base/test/vegetable/"

#搭建卷積神經網絡
from keras import layers
from keras import models
model=models.Sequential()
"""
Output shape計算公式：(輸入尺寸-卷積核尺寸/步長+1
對CNN模型，Param的計算方法如下：
卷積核長度*卷積核寬度*通道數+1）*卷積核個數
輸出圖片尺寸：150-3+1=148*148
參數數量：32*3*3*3+32=896
"""
model.add(layers.Conv2D(32,(3,3),activation='relu',input_shape=(150,150,3)))#卷積層1
model.add(layers.MaxPooling2D(2,2))#最大池化層1
# 輸出圖片尺寸：148/2=74*74 
# 輸出圖片尺寸：74-3+1=72*72，參數數量：64*3*3*32+64=18496
#32是第1個卷積層的輸出的通道數
model.add(layers.Conv2D(64, (3, 3), activation='relu'))#卷積層2
model.add(layers.MaxPooling2D((2, 2)))#最大池化層2
# 輸出圖片尺寸：72/2=36*36
 
#Output Shape的輸出為36
# 輸出圖片尺寸：36-3+1=34*34，參數數量：128*3*3*64+128=73856
model.add(layers.Conv2D(128, (3, 3), activation='relu'))#卷積層3
model.add(layers.MaxPooling2D((2, 2)))#最大池化層3
# 輸出圖片尺寸：34/2=17*17
# 輸出圖片尺寸：17-3+1=15*15，參數數量：128*3*3*128+128=147584
model.add(layers.Conv2D(128, (3, 3), activation='relu'))# 輸出圖片尺寸：15/2=7*7
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))   #全連接層1
model.add(layers.Dense(1, activation='sigmoid'))#全連接層2，作為輸出層
model.summary()

from keras import optimizers
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
               metrics=['acc'])

from keras.preprocessing.image import ImageDataGenerator
#歸一化

train_datagen = ImageDataGenerator(rescale=1./255)
validation_datagen=ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)
# 數據集加載函數，指明數據集的位置並統一處理為imgheight*imgwidth的大小，同時設置batch
    # 加載訓練集
train_ds = train_datagen.flow_from_directory (
        train_data_dir,
        class_mode='binary',
        target_size=(150, 150),
        batch_size=20)
    #加載測試集
test_ds = test_datagen.flow_from_directory(
        test_data_dir,
        class_mode='binary',
        target_size=(150, 150),
        batch_size=20)
    # 加載驗證集
val_ds =test_datagen.flow_from_directory(
        val_data_dir,
        class_mode='binary',
        target_size=(150, 150),
        batch_size=20)

for data_batch, labels_batch in train_ds:
    print('data batch shape:', data_batch.shape)
    print('labels batch shape:', labels_batch.shape)
    break

#訓練模型30輪次，可以修改epochs的值
history = model.fit_generator(
       train_ds,
       steps_per_epoch=100,
       epochs=30,
       validation_data=val_ds,
       validation_steps=50)

 #將訓練過程產生的數據保存為h5文件
model.save('fruit_and_vegetable_30epoch.h5')

import matplotlib.pyplot as plt
 
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(1, len(acc) + 1)

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure() 
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
 
plt.show()

from PIL import Image
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
#加載模型
model=load_model('fruit_and_vegetable_30epoch.h5')
#本地圖片路徑
img_path=("C:/Users/linyicheng/Desktop/base/train/fruit/apple_41.jpg")
img = image.load_img(img_path,target_size=(150,150))
# 將其轉換為具有形狀的Numpy數組（150、150、3）
img_tensor = image.img_to_array(img)/255
# 將其形態變為(150,150,150,3)的形狀
img_tensor = np.expand_dims(img_tensor, axis=0)
#取圖片信息
prediction =model.predict(img_tensor)
#輸出識別率
print(prediction)
if prediction>0.5:
    print('水果')
else:
    print('蔬菜')
plt.imshow(img)

from keras.preprocessing.image import ImageDataGenerator
from keras.models import load_model
from tensorflow.python.keras.preprocessing import image
from tensorflow.python.keras.preprocessing.image import ImageDataGenerator,array_to_img,img_to_array,load_img

import os
from keras.preprocessing.image import load_img
import matplotlib.pyplot as plt
figure,ax = plt.subplots(nrows=1, ncols=4, sharex=True, 
                         sharey=True,figsize=(16,10))

datagen=ImageDataGenerator(
    rotation_range=40,      #圖像隨機旋轉角度范圍
    width_shift_range=0.2,  #圖片在水平方向上平移的比例
    height_shift_range=0.2, #圖片在垂直方向上平移的比例
    shear_range=0.2,        #隨機錯切變換的角度
    zoom_range=0.2,         #圖像隨機縮放的范圍
    horizontal_flip=True,   #隨機將一半圖像水平翻轉
      
    fill_mode='nearest')
#填充創建像素的一種方法
fnames=[os.path.join(train_fruit_data,fname)for fname in os.listdir(train_fruit_data)]
img_path=fnames[6]
img=image.load_img(img_path,target_size=(150,150))
x = image.img_to_array(img)/255
x = x.reshape((1,) + x.shape)
i = 0
for batch in datagen.flow(x, batch_size=1):
    ax = ax.flatten()   # 將子圖從多維變成一維
    plt.figure(i)
    imgplot = ax[i-1].imshow(image.array_to_img(batch[0]), cmap='Greys', interpolation='nearest')
    i += 1
    if i % 4 == 0:
        break
plt.show() #繪制增強后的圖像
'''
雖然使用了數據增強技術，但是從輸出的圖像來看，圖片之間還是有很大的相似度，
因為它們均來自同一張原始圖片，並沒有提供新的信息。
為了盡可能消除過擬合，可以在模型中增加一個DDropout層，添加到密集連接分類
器前
'''
#在緊密連接的分類器之前為模型添加一個Dropout層
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

from keras import optimizers
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
               metrics=['acc'])

#使用數據增強和dropout來訓練我們的網絡
#歸一化
train_datagen = ImageDataGenerator(
     rescale=1./255,
     rotation_range=40,
     width_shift_range=0.2,
     height_shift_range=0.2,
     shear_range=0.2,
     zoom_range=0.2,
     horizontal_flip=True)
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
     train_data_dir,
     #輸入訓練圖像尺寸,所有圖片的size必須是150x150
     target_size=(150, 150),
     batch_size=32,
     #因為我們使用二元交叉熵損失，我們需要二元標簽
     class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
     val_data_dir,
     target_size=(150, 150),
     batch_size=32,
     class_mode='binary')

history = model.fit_generator(
       train_generator,
       steps_per_epoch=100,
       epochs=100,
       validation_data=validation_generator,
       validation_steps=50)

model.save('fruit_and_vegetable_100epoch.h5')

import matplotlib.pyplot as plt
 
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(1, len(acc) + 1)

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure() 
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
 
plt.show()

from keras.preprocessing import image

from keras.models import load_model
import numpy as np
#加載模型
model=load_model('fruit_and_vegetable_100epoch.h5')
#本地圖片路徑
img_path=("C:/Users/linyicheng/Desktop/base/train/fruit/apple_42.jpg")
img = image.load_img(img_path,target_size=(150,150))
# 將其轉換為具有形狀的Numpy數組（150、150、3）
img_tensor = image.img_to_array(img)/255
# 將其形態變為(150,150,150,3)的形狀
img_tensor = np.expand_dims(img_tensor, axis=0)
#取圖片信息
prediction =model.predict(img_tensor)
#輸出識別率
print(prediction)
if prediction>0.5:
    print('水果')
else:
    print('蔬菜')
plt.imshow(img)

from PIL import Image
from keras.preprocessing import image
import matplotlib.pyplot as plt
from keras.models import load_model
plt.rcParams['font.family'] = ['SimHei']
figure,ax = plt.subplots(nrows=2, ncols=4, sharex=True, sharey=True,figsize=(24,10))
def convertjpg(jpgfile):
    #將圖片縮小到（150，150）的大小
    img=Image.open(jpgfile)
    try:
        new_img=img.resize((150,150),Image.BILINEAR)
        return new_img
    except Exception as e:
        print(e)
snames=[os.path.join(test_vegetable_data,sname)for sname in os.listdir(test_vegetable_data)]
j=0
for i in snames[0:8] :
    j+=1
    img_show=convertjpg(i)
    img_scale = image.img_to_array(img_show)
    img_scale = img_scale.reshape(1,150,150,3) # 將形狀轉化為(1,150,150,3)
    img_scale = img_scale.astype('float32')/255
    result = model.predict(img_scale)# 預測函數-
    if result>0.5:
        title = '蔬菜'
        label = '正確識別率為：'+str(result)
    else:
        title = '水果'
        label = '正確識別率為：'+str(1-result)
         
    ax = ax.flatten()   # 將子圖從多維變成一維
    ax[j-1].imshow(img_show, cmap='Greys', interpolation='nearest')
     # 子圖標題
    ax[j-1].set_title(title,fontsize=24)
     # 子圖X軸標簽
    ax[j-1].set_xlabel(label,fontsize=24)
     # 去掉刻度
    ax[0].set_xticks([])
    ax[0].set_yticks([]) 
plt.show()

figure,ax = plt.subplots(nrows=2, ncols=4, sharex=True, sharey=True,figsize=(24,10))

fnames=[os.path.join(test_fruit_data,fname)for fname in os.listdir(test_fruit_data)]
j=0
for i in fnames[0:8]: 
    j+=1
    img_show=convertjpg(i)
    img_scale = image.img_to_array(img_show)
    img_scale = img_scale.reshape(1,150,150,3) # 將形狀轉化為(1,150,150,3)
    img_scale = img_scale.astype('float32')/255
    result = model.predict(img_scale)# 預測函數-
    if result>0.5:
        title = '蔬菜'
        label = '正確識別率為：'+ str(result)
    else:
        title = '水果'
        label = '正確識別率為：'+str(1-result)         
    ax = ax.flatten()   # 將子圖從多維變成一維
    ax[j-1].imshow(img_show, cmap='Greys', interpolation='nearest')
     # 子圖標題
    ax[j-1].set_title(title,fontsize=24)
     # 子圖X軸標簽
    ax[j-1].set_xlabel(label,fontsize=24)
    # 去掉刻度
    ax[0].set_xticks([])
    ax[0].set_yticks([])

四、總結

此次機器學習過程中參考了minist案例和貓狗識別案例，在數據處理構建數據集目錄是出現一些問題，開始思路沒有構建好，導致數據處理花費太多時間，以及代碼太過冗長，還有就是在導入一些必要的數據庫是出現問題如PIL，在網上查明原因說需要更新pillow和PIL的版本，后續應多加練習熟悉所需的各種環境。

這次機器學習中，無論是第一次訓練模型還是增強后的模型訓練都達到了預期的效果，但是在進行機器學習設計的過程中發現自己對程序設計的思路沒有理清楚，導致花費了大量時間改正，以后在進行類似學習前應當將思路理清，再付諸行動。