TA11 画像分類 検討

#インポート
# import tensorflow
import keras
from PIL import Image
import glob
# import os
import numpy as np
# from PIL import ImageFile

import matplotlib.pyplot as plt

from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense

from keras.utils import np_utils
#収束判定停止用
from keras.callbacks import EarlyStopping
#オプティマイザー
from tensorflow.keras.optimizers import RMSprop
# from tensorflow.keras.optimizers import SGD
# from tensorflow.keras.optimizers import Adam

#-------パラメータ-------

#>>>>>>>>
#データ区分け
folder = ["Cantilever2",  "Double_sided2"] 
#作成npy名
modelName = "beamtest_V4"
saveModel = "./model/" + modelName #h5ファイル
#>>>>>>>>

image_size = 64
loss = 'binary_crossentropy'
num_folder = len(folder)

batch_size = 10

epochs = 50
nomalize1 = 255.0

# EaelyStoppingの設定
stopFlag = 1
early_stopping =  EarlyStopping(
                            monitor='loss',
                            min_delta=0.00001,
                            patience=4,
                    )

#計算ログ出力
# keras.callbacks.CSVLogger(filename, separator=',', append=False)
csv_logger = keras.callbacks.CSVLogger(modelName + '_training.log')
#------------------------

#宣言
X_train = []
X_test  = []
y_train = []
y_test  = []

#学習データ編集
for index, classlabel in enumerate(folder):
    photos_dir = "./" + classlabel
    files = glob.glob(photos_dir + "/*")

    for i, file in enumerate(files):
        image = Image.open(file)
        image = image.convert("RGB")
        image = image.resize((image_size, image_size))
        data = np.asarray(image)
        #格納
        X_test.append(data)
        y_test.append(index)
        
        #データ拡張　回転18step
        for angle in range(0, 360, 20):
            img_r = image.rotate(angle)
            data = np.asarray(img_r)
            X_train.append(data)
            y_train.append(index)

            #左右反転
            img_trains = image.transpose(Image.FLIP_LEFT_RIGHT)
            data = np.asarray(img_trains)
            X_train.append(data)
            y_train.append(index)

            #上下反転
            img_trains = image.transpose(Image.FLIP_TOP_BOTTOM)
            data = np.asarray(img_trains)
            X_train.append(data)
            y_train.append(index)

#格納
X_train = np.array(X_train)
X_test  = np.array(X_test)
y_train = np.array(y_train)
y_test  = np.array(y_test)

xydata = (X_train, X_test, y_train, y_test)
np.save("./" + saveModel + ".npy" , xydata)

#------------------------
#データ読み込み
#------------------------
def load_data():
    X_train, X_test, y_train, y_test = np.load("./" + saveModel + ".npy", allow_pickle=True)
    # 0-1に正規化
    X_train = X_train.astype("float") / nomalize1
    X_test  = X_test.astype("float") / nomalize1
    y_train = np_utils.to_categorical(y_train, num_folder)
    y_test  = np_utils.to_categorical(y_test, num_folder)

    return X_train, y_train, X_test, y_test

#------------------------
#学習の実施
#------------------------
def train(X, y, X_test, y_test):

    model = Sequential([
        Conv2D(32,(3,3), padding='valid',input_shape=X.shape[1:]),
        Activation('relu'),
        Conv2D(32,(3,3)),
        Activation('relu'),
        MaxPooling2D(pool_size=(2,2)),
        Dropout(0.1),

        Conv2D(64,(3,3), padding='valid'),
        Activation('relu'),
        Conv2D(64,(3,3)),
        Activation('relu'),
        MaxPooling2D(pool_size=(2,2)),
        Dropout(0.25),

        Flatten(),
        Dense(512),
        Activation('relu'),
        Dropout(0.45),
        Dense(2),
        Activation('softmax')
    ])
 
    # opt = RMSprop(lr=0.001, rho=0.9, epsilon=None, decay=0.0)


    model.compile(loss = loss,optimizer=RMSprop(lr=0.001, rho=0.9, epsilon=None, decay=0.0),metrics=['accuracy'])
    print(model.loss)
    if stopFlag == 1:
        model.fit(X, y, batch_size=batch_size, epochs=epochs, callbacks=[early_stopping,csv_logger]) # 途中停止
    else:
        model.fit(X, y, batch_size=batch_size, epochs=epochs, callbacks=[csv_logger]) 

    model.save('./' + saveModel +'.h5')    # HDF5ファイルにKerasのモデルを保存
    # model.save('./'+ saveModel) 　#モデルセット作成
    return model


#------------------------
#データ読み込みと学習の実施
#------------------------
def main():
    # データの読み込み
    X_train, y_train, X_test, y_test = load_data()
    # モデルの学習
    train(X_train, y_train, X_test, y_test)
main()

import cv2 as cv2
import numpy as np
import os
from keras.models import load_model
import glob
from PIL import Image

#-------パラメータ-------
#学習モデルの読み込み
saveModel = './model/' + "beamtest_V4.h5"
 
imgsize = (64, 64) 
nomalize1 = 255.0
judgeStr = ["Cantilever",  "Double_sided"]


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

model = load_model( saveModel)

#画像整理関数
def check_image(path):
    img = Image.open(path)
    img = img.convert('RGB')
    img = img.resize(imgsize)
    img = np.asarray(img)
    img = img / nomalize1
    return img

count=0
for i in files_list:

    img = check_image(path+files_list[count])

    print(":"+files_list[count])
    print(model.predict(np.array([img]))) # 精度の表示
    judeassy = np.argmax(model.predict(np.array([img])), axis=1)
    if judeassy == 0:
        print("[判定] " + judgeStr[0])
    elif judeassy == 1:
        print("[判定] "+ judgeStr[1])
    count+=1