深度學習入門（2）----用卷積神經網絡進行圖像識別(二)

在深度學習入門（1）----用卷積神經網絡進行圖像識別(一)中已經詳細介紹了卷積神經網絡網絡進行圖像識別的基本應用，本文在上文的基礎上，主要解決如下的問題：
實現一個包含4個卷積層、2個全連接層的卷積神經網絡來處理88點陣的灰度圖像識別。
說明：請自行準備一個包含灰度像素的文本文件checkData64.txt，其中每行前64個數字是88矩陣中每個點的灰度，灰度值爲0-255，最後3個數字還是3種情況的概率，即每行67個數字。
代碼如下：

#!/usr/bin/env python 
# -*- coding:utf-8 -*-
import tensorflow as tf
import numpy as np
import pandas as pd
import sys


roundCount = 100
learnRate = 0.01

argt = sys.argv[1:]

for v in argt:
    if v.startswith("-round="):
        roundCount = int(v[len("-round="):])
    if v.startswith("-learnrate="):
        learnRate = float(v[len("-learnrate="):])


fileData = pd.read_csv('checkData64.txt', dtype=np.float32, header=None)

wholeData = fileData.as_matrix()

rowCount = wholeData.shape[0]

print("wholeData=%s" % wholeData)
print("rowSize=%d" % wholeData.shape[1])
print("rowCount=%d" % rowCount)

x = tf.placeholder(shape=[64], dtype=tf.float32)
yTrain = tf.placeholder(shape=[3], dtype=tf.float32)

filter1T = tf.Variable(tf.ones([2, 2, 1, 1]), dtype=tf.float32)

n1 = tf.nn.conv2d(input=tf.reshape(x, [1, 8, 8, 1]), filter=filter1T, strides=[1, 1, 1, 1], padding='SAME')

filter2T = tf.Variable(tf.ones([2, 2, 1, 1]), dtype=tf.float32)

n2 = tf.nn.conv2d(input=tf.reshape(n1, [1, 8, 8, 1]), filter=filter2T, strides=[1, 1, 1, 1], padding='VALID')

filter3T = tf.Variable(tf.ones([2, 2, 1, 1]), dtype=tf.float32)

n3 = tf.nn.conv2d(input=tf.reshape(n2, [1, 7, 7, 1]), filter=filter3T, strides=[1, 1, 1, 1], padding='VALID')

filter4T = tf.Variable(tf.ones([2, 2, 1, 1]), dtype=tf.float32)

n4 = tf.nn.conv2d(input=tf.reshape(n3, [1, 6, 6, 1]), filter=filter4T, strides=[1, 1, 1, 1], padding='VALID')

n4f = tf.reshape(n4, [1, 25])

w4 = tf.Variable(tf.random_normal([25, 32]), dtype=tf.float32)
b4 = tf.Variable(0, dtype=tf.float32)

n4 = tf.nn.tanh(tf.matmul(n4f, w4) + b4)

w5 = tf.Variable(tf.random_normal([32, 3]), dtype=tf.float32)
b5 = tf.Variable(0, dtype=tf.float32)

n5 = tf.reshape(tf.matmul(n4, w5) + b5, [-1])

y = tf.nn.softmax(n5)

loss = -tf.reduce_mean(yTrain * tf.log(tf.clip_by_value(y, 1e-10, 1.0)))
optimizer = tf.train.RMSPropOptimizer(learnRate)

train = optimizer.minimize(loss)

sess = tf.Session()
sess.run(tf.global_variables_initializer())

for i in range(roundCount):
    lossSum = 0.0

    for j in range(rowCount):
        result = sess.run([train, x, yTrain, y, loss], feed_dict={x: wholeData[j][0:64], yTrain: wholeData[j][64:67]})

        lossT = float(result[len(result) - 1])

        lossSum = lossSum + lossT

        if j == (rowCount - 1):
            print("i: %d, loss: %10.10f, avgLoss: %10.10f" % (i, lossT, lossSum / (rowCount + 1)))