Omniglot数据集分类器的设计与实现——TensorFlow期末考核

from skimage import io, transform # skimage模块下的io transform(图像的形变与缩放)模块 import glob # glob 文件通配符模块 import os # os 处理文件和目录的模块 import tensorflow as tf import numpy as np # 多维数据处理模块 import time # 数据集地址 path = 'data' # 模型保存地址 model_path = 'model/fc_model.ckpt' # 将所有的图片resize成100*100 w = 100 h = 100 c = 1 # 读取图片+数据处理 def read_img(path): imgs = [] labels = [] for fistname in os.listdir(path): # print(fistname) for textname in os.listdir(path + "/" + fistname): # os.listdir() 方法用于返回指定的文件夹包含的文件或文件夹的名字的列表 # print(textname) for fname in os.listdir(path + "/" + fistname + "/" + textname): # print(fname) fpath = os.path.join(path, fistname, textname, fname) # 路径拼接 img = io.imread(fpath) # 按照路径打开文件 # #skimage.transform.resize(image, output_shape)改变图片的尺寸 img = transform.resize(img, (w, h)) # 将读取的图片数据加载到imgs[]列表中 imgs.append(img) # 将图片的label加载到labels[]中，与上方的imgs索引对应 label = int(fistname[-1]) labels.append(label) # 将读取的图片和labels信息，转化为numpy结构的ndarr(N维数组对象（矩阵）)数据信息 imgs = np.array(imgs) imgs = np.reshape(imgs, [-1, w, h, 1]) print("shape of datas: {}\tshape of labels: {}".format(np.array(imgs).shape, np.array(labels).shape)) return np.asarray(imgs, np.float32), np.asarray(labels, np.int32) # 调用读取图片的函数，得到图片和labels的数据集 data, label = read_img(path) # 打乱顺序 # 读取data矩阵的第一维数（图片的个数） num_example = data.shape[0] # 产生一个num_example范围，步长为1的序列 arr = np.arange(num_example) # 调用函数，打乱顺序 np.random.shuffle(arr) # 按照打乱的顺序，重新排序 data = data[arr] label = label[arr] # 将所有数据分为训练集和验证集 ratio = 0.8 s = np.int(num_example * ratio) x_train = data[:s] y_train = label[:s] x_val = data[s:] y_val = label[s:] # -----------------构建网络---------------------- # 本程序cnn网络模型，共有7层，前三层为卷积层，后三层为全连接层，前三层中，每层包含卷积、激活、池化层 # 占位符设置输入参数的大小和格式 x = tf.placeholder(tf.float32, shape=[None, w, h, c], name='x') y_ = tf.placeholder(tf.int32, shape=[None, ], name='y_') def inference(input_tensor, train, regularizer): # -----------------------第一层---------------------------- with tf.variable_scope('layer1-conv1'): # 初始化权重conv1_weights为可保存变量，大小为5x5,3个通道（RGB），数量为32个 conv1_weights = tf.get_variable("weight", [5, 5, 1, 32], initializer=tf.truncated_normal_initializer(stddev=0.1)) # 初始化偏置conv1_biases，数量为32个 conv1_biases = tf.get_variable("bias", [32], initializer=tf.constant_initializer(0.0)) # 卷积计算，tf.nn.conv2d为tensorflow自带2维卷积函数，input_tensor为输入数据， # conv1_weights为权重，strides=[1, 1, 1, 1]表示左右上下滑动步长为1，padding='SAME'表示输入和输出大小一样，即补0 conv1 = tf.nn.conv2d(input_tensor, conv1_weights, strides=[1, 1, 1, 1], padding='SAME') # 激励计算，调用tensorflow的relu函数 relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_biases)) with tf.name_scope("layer2-pool1"): # 池化计算，调用tensorflow的max_pool函数，strides=[1,2,2,1]，表示池化边界，2个对一个生成，padding="VALID"表示不操作。 pool1 = tf.nn.max_pool(relu1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="VALID") # -----------------------第二层---------------------------- with tf.variable_scope("layer3-conv2"): # 同上，不过参数的有变化，根据卷积计算和通道数量的变化，设置对应的参数 conv2_weights = tf.get_variable("weight", [5, 5, 32, 64], initializer=tf.truncated_normal_initializer(stddev=0.1)) conv2_biases = tf.get_variable("bias", [64], initializer=tf.constant_initializer(0.0)) conv2 = tf.nn.conv2d(pool1, conv2_weights, strides=[1, 1, 1, 1], padding='SAME') relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_biases)) with tf.name_scope("layer4-pool2"): pool2 = tf.nn.max_pool(relu2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID') # -----------------------第三层---------------------------- # 同上，不过参数的有变化，根据卷积计算和通道数量的变化，设置对应的参数 with tf.variable_scope("layer5-conv3"): conv3_weights = tf.get_variable("weight", [3, 3, 64, 128], initializer=tf.truncated_normal_initializer(stddev=0.1)) conv3_biases = tf.get_variable("bias", [128], initializer=tf.constant_initializer(0.0)) conv3 = tf.nn.conv2d(pool2, conv3_weights, strides=[1, 1, 1, 1], padding='SAME') relu3 = tf.nn.relu(tf.nn.bias_add(conv3, conv3_biases)) with tf.name_scope("layer6-pool3"): pool3 = tf.nn.max_pool(relu3, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID') # -----------------------第四层---------------------------- # 同上，不过参数的有变化，根据卷积计算和通道数量的变化，设置对应的参数 with tf.variable_scope("layer7-conv4"): conv4_weights = tf.get_variable("weight", [3, 3, 128, 128], initializer=tf.truncated_normal_initializer(stddev=0.1)) conv4_biases = tf.get_variable("bias", [128], initializer=tf.constant_initializer(0.0)) conv4 = tf.nn.conv2d(pool3, conv4_weights, strides=[1, 1, 1, 1], padding='SAME') relu4 = tf.nn.relu(tf.nn.bias_add(conv4, conv4_biases)) with tf.name_scope("layer8-pool4"): pool4 = tf.nn.max_pool(relu4, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID') nodes = 6 * 6 * 128 reshaped = tf.reshape(pool4, [-1, nodes]) # 使用变形函数转化结构 # -----------------------第五层--------------------------- with tf.variable_scope('layer9-fc1'): # 初始化全连接层的参数，隐含节点为1024个 fc1_weights = tf.get_variable("weight", [nodes, 1024], initializer=tf.truncated_normal_initializer(stddev=0.1)) if regularizer != None: tf.add_to_collection('losses', regularizer(fc1_weights)) # 正则化矩阵 fc1_biases = tf.get_variable("bias", [1024], initializer=tf.constant_initializer(0.1)) # 使用relu函数作为激活函数 fc1 = tf.nn.relu(tf.matmul(reshaped, fc1_weights) + fc1_biases) # 采用dropout层，减少过拟合和欠拟合的程度，保存模型最好的预测效率 if train: fc1 = tf.nn.dropout(fc1, 0.5) # -----------------------第六层---------------------------- with tf.variable_scope('layer10-fc2'): # 同上，不过参数的有变化，根据卷积计算和通道数量的变化，设置对应的参数 fc2_weights = tf.get_variable("weight", [1024, 512], initializer=tf.truncated_normal_initializer(stddev=0.1)) if regularizer != None: tf.add_to_collection('losses', regularizer(fc2_weights)) fc2_biases = tf.get_variable("bias", [512], initializer=tf.const