TA12 画像分類2 検討

# -*- coding: utf-8 -*-
import numpy as np
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D
from keras.optimizers import RMSprop
from keras.preprocessing.image import img_to_array, load_img
from keras.utils import np_utils
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
import os
import pickle
import glob

#------------------------
#パラメータ
#------------------------
#ファイルフォルダ
fileDir = os.path.dirname(__file__)
# ルートパス
rootPathTRAIN = fileDir +"\\"
# 作成物フォルダ
res_PATH = rootPathTRAIN + "LearnModel\\"

#データ区分け　参照フォルダ
Part = ["Cantilever2","Double_sided2","fluid"]
num_Part = len(Part)

# 入力画像のパラメータ
pic_width = 32 # 入力画像の幅
pic_height = 32 # 入力画像の高さ
pic_ch = 3 # 3ch画像（RGB）で学習
pic_size = 64

#作成データ
makeModel = "model.json"
weightData = "weightData.hdf5"
learnHis = "learnHis.json"
leanGraph = "leanGraph.png"

#学習パラメータ
batchSize = 5 # バッチサイズ
numClasses = num_Part # クラス数
epochs = 50      # エポック数
dropoutRate = 0.2 # 過学習防止にデータ20%を削除
nomalize1 = 255.0

#学習確認グラフ
graphSize = 12
graphHeigt = 10
graphFont = 25



#データの格納
# catNum = 0
# imgCount = 0
# picData =[]

def plotHIS(history, 
                checkGraphPath, 
                picSizeWidth, 
                piSsizeHeight, 
                fontSize):

    acc = history.history['accuracy']
    val_acc = history.history['val_accuracy']
    loss = history.history['loss']
    val_loss = history.history['val_loss']
   
    epochs = range(len(acc))

    plt.figure(figsize=(picSizeWidth, piSsizeHeight))
    plt.rcParams['font.family'] = 'Times New Roman'
    plt.rcParams['font.size'] = fontSize
    
    plt.plot(epochs, acc, color = "blue", linestyle = "solid", label = 'train acc')
    plt.plot(epochs, val_acc, color = "green", linestyle = "solid", label= 'valid acc')
    
    plt.plot(epochs, loss, color = "red", linestyle = "solid" ,label = 'train loss')
    plt.plot(epochs, val_loss, color = "orange", linestyle = "solid" , label= 'valid loss')
    
    plt.legend()
    plt.grid()

    plt.savefig(checkGraphPath)
    plt.close() 

def main():


    data_x = []
    data_y = []
    data_x_pic = []
    data_y_pic = []
    numClasses = num_Part

    for index, cateP in enumerate(Part):
        photos_dir = rootPathTRAIN + cateP
        files = glob.glob(photos_dir + "/*")

#-----------------------------------------------------
        for i,filepath in enumerate(files):
        # for filepath in imgDaras(rootPathTRAIN +"" +cateP):
            img = img_to_array(load_img(filepath, target_size=(pic_width,pic_height, pic_ch)))
            data_x.append(img)
            data_y.append(index) # 教師データ（正解）

                
        # #画像の書き出し
        # imgCount = 0
        # for datPic in picData:
        #     imgCount = imgCount + 1
        #     datPic.save(res_PATH +'lena_flip'+'_'+str(imgCount)+'.jpg', quality=95)

#-----------------------------------------------------

    data_x = np.asarray(data_x) # NumPy配列

    data_y = np.asarray(data_y) # NumPy配列

    #学習用と検査用にデータを分割する
    x_train, x_test, y_train, y_test = train_test_split(data_x, data_y, test_size=0.2)

    x_train = x_train.astype('float')
    x_test = x_test.astype('float')
    x_train = x_train / nomalize1
    x_test = x_test / nomalize1

    # ラベルのコーディング
    y_train = np_utils.to_categorical(y_train, numClasses)
    y_test = np_utils.to_categorical(y_test, numClasses)

    # データ個数
    print(x_train.shape, 'x train samples')
    print(x_test.shape, 'x test samples')
    print(y_train.shape, 'y train samples')
    print(y_test.shape, 'y test samples')

    #モデル作成
    model = Sequential()
    #2次元畳み込み層
    model.add(Conv2D(32,(3,3), 
                padding='same', 
                input_shape=x_train.shape[1:],
                activation='relu'))
    #2次元畳み込み層
    model.add(Conv2D(32,(3,3),
                padding='same',
                activation='relu'))
    #プーリング層
    model.add(MaxPooling2D(pool_size=(2, 2)))
    # ドロップアウト
    model.add(Dropout(dropoutRate))
    #2次元畳み込み層
    model.add(Conv2D(64,(3,3),
                padding='same',
                activation='relu'))
    #2次元畳み込み層
    model.add(Conv2D(64,(3,3),
                padding='same',
                activation='relu'))
    #プーリング層
    model.add(MaxPooling2D(pool_size=(2, 2)))
    # ドロップアウト
    model.add(Dropout(dropoutRate))
    #次元削減
    model.add(Flatten())
    # 全結合層
    model.add(Dense(512, activation='relu'))
    # ドロップアウト
    model.add(Dropout(dropoutRate))
    # 全結合層
    model.add(Dense(numClasses, activation='softmax')) # 活性化関数：softmax


    # モデル表示
    #model.summary()

    # コンパイル
    model.compile(loss='categorical_crossentropy', optimizer=RMSprop(lr=0.001, rho=0.9, epsilon=None, decay=0.0), metrics=['accuracy'])

    #モデルの学習　データの10%を使用して検証
    history = model.fit(x_train, 
                        y_train, 
                        batch_size=batchSize, 
                        epochs=epochs, 
                        # verbose=1, 
                        validation_split=0.1)

    # テスト用データで確認
    check = model.evaluate(x_test, 
                            y_test,
                            verbose=0
                            )


    # 損失値確認
    print('check loss:', check[0])

    # 正解率
    print('check accuracy:', check[1])
    
    # 学習過程をグラフ
    plotHIS(history, 
                checkGraphPath = res_PATH + leanGraph, 
                picSizeWidth = graphSize, 
                piSsizeHeight = graphHeigt, 
                fontSize = graphFont)

    #構造セーブ
    open(res_PATH  + makeModel,"w").write(model.to_json())  

    #学習の重み保存
    model.save_weights(res_PATH + weightData)

    #履歴保存
    with open(res_PATH + learnHis, 'wb') as f:
        pickle.dump(history.history, f)


if __name__ == '__main__':

    #ファイル削除
    if(os.path.isfile(res_PATH + leanGraph)):
        os.remove(res_PATH + leanGraph)
    if(os.path.isfile(res_PATH + makeModel)):
        os.remove(res_PATH + makeModel)
    if(os.path.isfile(res_PATH + weightData)):
        os.remove(res_PATH + weightData)
    if(os.path.isfile(res_PATH + learnHis)):
        os.remove(res_PATH + learnHis)
    
    main()

# -*- coding: utf-8 -*-
import numpy as np
from keras.models import model_from_json
from keras.preprocessing.image import load_img, img_to_array
import glob
import os

#------------------------
#パラメータ
#------------------------

# ラベル
labels =['片持ち梁', '両持ち梁', '流体']

#ファイルフォルダ
fileDir = os.path.dirname(__file__)

# ルートパス
rootPathTRAIN = fileDir +"\\"

#学習データ保存フォルダ
lernPath = rootPathTRAIN + "LearnModel\\"

#検査するファイル
checkFile = rootPathTRAIN + "judge\\"

#評価対象ファイル
files_list=[]
path =  rootPathTRAIN + "judge\\"
files = glob.glob(path + "/*")
for f in files:
    files_list.append(os.path.basename(f))

#作成データ
makeModel = "model.json"
weightData = "weightData.hdf5"

# 画像情報
picWidth = 32
picHeight = 32
colorCh = 3

nomalize1 = 255.0

#-------------------処理----------------------------
#モデルの読み込み
model = model_from_json(open(lernPath + makeModel).read())

#重みを読み込み
model.load_weights(lernPath + weightData)

count=0
for i in files_list:

    img = load_img(checkFile + i, target_size=(picWidth, picHeight))
    img = img_to_array(img) 
    img = img.astype('float')/nomalize1
    img = np.array([img])

    # データ予測実行
    y_pred = model.predict(img)

    # 要素番号確定
    number_pred = np.argmax(y_pred) 

    # 予測表示
    print("#---------------")
    print("ファイル名" +i)
    print("index:", number_pred)  # ラベル番号
    print('[判定]:', labels[int(number_pred)]) # 予想ラベル
    print("predict:", y_pred)  # 出力値