本文共 8591 字，大约阅读时间需要 28 分钟。
通过可以训练得到模型，并将模型进行保存。本次博文中主要是调用训练好的进行预测。
该模型训练所使用的数据集是MNIST集。在对图片中数字进行识别前，需要对图片进行预处理，即将图片转变为MNIST数据集中图片的格式。预处理程序可以调用模型进行预测的代码如下：
import tensorflow as tffrom tensorflow.examples.tutorials.mnist import input_dataimport cv2import numpy as np#import imutils# from scikit-image import data,segmentation,measure,morphology,colorfrom PIL import Imageim = Image.open('D:\\deng\\ppp\\888.png')data = list(im.getdata())result = [(255-x)*1.0/255.0 for x in data]print(result)sess = tf.Session()def compute_accuracy(v_xs, v_ys): global prediction y_pre = sess.run(prediction, feed_dict={ xs: v_xs, keep_prob: 1}) #(10000,10) correct_prediction = tf.equal(tf.argmax(y_pre, 1), tf.argmax(v_ys, 1)) #比较是否相等，返回bool accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) #将比较结果转换成tf.float32，并计算平均值 result = sess.run(accuracy, feed_dict={ xs: v_xs, ys: v_ys, keep_prob: 1}) return resultdef weight_variable(shape): initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial)def bias_variable(shape): initial = tf.constant(0.1, shape=shape) return tf.Variable(initial)def conv2d(x, W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')def max_pool_2x2(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')keep_prob = tf.placeholder(tf.float32)with tf.name_scope('inputs'): xs = tf.placeholder(tf.float32, [None, 784], name='x_input') #28x28 ys = tf.placeholder(tf.float32, [None, 10], name='y_input')x_image = tf.reshape(xs, [-1, 28, 28, 1])#print("n_samples:", x_image.shape)#conv1 layerW_conv1 = weight_variable([5, 5, 1, 32]) #patch 5x5 in size=1, out size 32b_conv1 = bias_variable([32])h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1) # output size 28x28x32h_pool1 = max_pool_2x2(h_conv1) # size 14x14x32#conv2 layerW_conv2 = weight_variable([5, 5, 32, 64]) #patch 5x5 in size=32, out size 64b_conv2 = bias_variable([64])h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) # output size 14x14x64h_pool2 = max_pool_2x2(h_conv2) # size 7x7x64#func1 layerW_fc1 = weight_variable([7*7*64, 1024])b_fc1 = bias_variable([1024])h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)#func2 layerW_fc2 = weight_variable([1024, 10])b_fc2 = bias_variable([10])prediction = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction), reduction_indices=[1])) #losstf.summary.scalar('loss', cross_entropy)with tf.name_scope('train'): train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)init = tf.global_variables_initializer()#summary writer goes in heremerged = tf.summary.merge_all()train_writer = tf.summary.FileWriter('D:/deng/logs/train', sess.graph)test_writer = tf.summary.FileWriter('D:/deng/logs/test', sess.graph)sess.run(init)saver = tf.train.Saver()#与训练过程代码进行对比，主要不同的地方在这里with tf.Session() as sess: sess.run(tf.global_variables_initializer()) saver.restore(sess, "D:/deng/model/model.ckpt")#这里使用了之前保存的模型参数 prediction = tf.argmax(prediction, 1) predint = prediction.eval(feed_dict={ xs: [result], keep_prob: 1.0}, session=sess) print("recognize result: %d" %predint[0])sess.close() 这个代码实测有效，只不过代码太过冗长了。本人会继续学习，尝试以最短的代码调用训练好的模型进行预测。若有什么建议或者错误之处，还请看到博文的朋友能够在评论区指出。感谢！
参考：
方法2：
相对于方法1，更加强大，并且简洁。感觉太爽了~import tensorflow as tffrom PIL import Imageim = Image.open('D:\\deng\\ppp\\333.png')data = list(im.getdata())result = [(255-x)*1.0/255.0 for x in data]print(result)with tf.Session() as sess: sess.run(tf.global_variables_initializer()) saver = tf.train.import_meta_graph('D:/deng/model2/model.ckpt.meta') saver.restore(sess, "D:/deng/model2/model.ckpt") # 这里使用了之前保存的模型参数 pred = tf.get_collection('network-output')[0] prediction = tf.argmax(pred, 1) graph = tf.get_default_graph() xs = graph.get_operation_by_name('x_input').outputs[0] keep_prob = graph.get_operation_by_name('keep_prob').outputs[0] #keep_prob = graph.get_operation_by_name('y_inout').outputs[0] predint = prediction.eval(feed_dict={ xs: [result], keep_prob: 1.0}, session=sess) print("recognize result: %d" % predint[0]) 训练模型的代码如下：
import tensorflow as tffrom tensorflow.examples.tutorials.mnist import input_datamnist = input_data.read_data_sets('MNIST_data', one_hot=True)sess = tf.Session()def compute_accuracy(v_xs, v_ys): global prediction y_pre = sess.run(prediction, feed_dict={ xs: v_xs, keep_prob: 1}) correct_prediction = tf.equal(tf.argmax(y_pre, 1), tf.argmax(v_ys, 1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) result = sess.run(accuracy, feed_dict={ xs: v_xs, ys: v_ys, keep_prob: 1}) return resultdef weight_variable(shape): initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial)def bias_variable(shape): initial = tf.constant(0.1, shape=shape) return tf.Variable(initial)def conv2d(x, W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')def max_pool_2x2(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')keep_prob = tf.placeholder(tf.float32, name='keep_prob')# with tf.name_scope('inputs'):xs = tf.placeholder(tf.float32, [None, 784], name='x_input')ys = tf.placeholder(tf.float32, [None, 10], name='y_input')x_image = tf.reshape(xs, [-1, 28, 28, 1])#print("n_samples:", x_image.shape)#conv1 layerwith tf.name_scope('conv1_layer'): with tf.name_scope('W_conv1'): W_conv1 = weight_variable([5, 5, 1, 32]) tf.summary.histogram('conv1/wights', W_conv1) with tf.name_scope('b_conv1'): b_conv1 = bias_variable([32]) tf.summary.histogram('conv1/biases', b_conv1) with tf.name_scope('conv1-wx_plus_b'): h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1) with tf.name_scope('conv1_pooling'): h_pool1 = max_pool_2x2(h_conv1)#conv2 layerwith tf.name_scope('conv2_layer'): with tf.name_scope('W_conv2'): W_conv2 = weight_variable([5, 5, 32, 64]) tf.summary.histogram('conv2/wights', W_conv2) with tf.name_scope('b_conv2'): b_conv2 = bias_variable([64]) tf.summary.histogram('conv2/biases', b_conv2) with tf.name_scope('conv2-wx_plus_b'): h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) with tf.name_scope('conv2_pooling'): h_pool2 = max_pool_2x2(h_conv2)#func1 layerwith tf.name_scope('full-connected1'): with tf.name_scope('W_fc1'): W_fc1 = weight_variable([7*7*64, 1024]) tf.summary.histogram('fc1/wights', W_fc1) with tf.name_scope('b_fc1'): b_fc1 = bias_variable([1024]) tf.summary.histogram('fc1/biases', b_fc1) with tf.name_scope('fc1-wx_plus_b'): h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64]) h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1) h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)#func2 layerwith tf.name_scope('full-connected2'): with tf.name_scope('W_fc2'): W_fc2 = weight_variable([1024, 10]) tf.summary.histogram('fc2/wights', W_fc2) with tf.name_scope('b_fc2'): b_fc2 = bias_variable([10]) tf.summary.histogram('fc2/biases', b_fc2) with tf.name_scope('fc2-wx_plus_b'): prediction = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2) tf.add_to_collection('network-output', prediction)cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction), reduction_indices=[1]))tf.summary.scalar('loss', cross_entropy)with tf.name_scope('train'): train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)init = tf.global_variables_initializer()#summary writer goes in heremerged = tf.summary.merge_all()train_writer = tf.summary.FileWriter('D:/deng/logs/train', sess.graph)#test_writer = tf.summary.FileWriter('D:/deng/logs/test', sess.graph)sess.run(init)saver = tf.train.Saver()for i in range(2000): batch_xs, batch_ys = mnist.train.next_batch(100) sess.run(train_step, feed_dict={ xs: batch_xs, ys: batch_ys, keep_prob: 0.5}) if i % 50 == 0: result = sess.run(merged, feed_dict={ xs: batch_xs, ys: batch_ys, keep_prob: 1}) train_writer.add_summary(result, i) print(compute_accuracy( mnist.test.images, mnist.test.labels ))saver = tf.train.Saver()model_path = "D:/deng/model2/model.ckpt"saver.save(sess, model_path)sess.close() 参考：
转载地址：http://rsfef.baihongyu.com/