全连接神经网络手写体推理代码实现--基于python tensorflow

一、介绍
输入为28x28的图片的浮点型数据
输出0-9的概率值

二、代码

#导入tensorflow库
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import numpy as np


#下载数据集
mnist = input_data.read_data_sets('MNIST_data/', one_hot=True)


#输入
x = tf.placeholder(tf.float32, [None, 784])

#目标输出
y = tf.placeholder(tf.float32, [None, 10])

#第一层实现
w1 = tf.Variable(tf.truncated_normal([784,64]), dtype = tf.float32)
b1 = tf.Variable(tf.zeros([1,64]), dtype = tf.float32)

a1 = tf.nn.relu(tf.matmul(x, w1) + b1)

#第二层实现
w2 = tf.Variable(tf.truncated_normal([64,10]), dtype = tf.float32)
b2 = tf.Variable(tf.zeros([1,10]), dtype = tf.float32)

y_ = tf.nn.softmax(tf.matmul(a1,w2) + b2)

#计算偏差 代入损失函数 梯度下降
#损失函数
loss = tf.reduce_mean(-tf.reduce_sum(y*tf.log(y_), axis = 1))


#梯度下降
train_step = tf.train.AdamOptimizer(0.0001).minimize(loss)

#初始化操作
#初始化参数节点操作
init = tf.global_variables_initializer()
#启动会话
sess = tf.Session()
sess.run(init)

#定义损失值和正确的识别率
#损失=实际输出与目标的差值
correct_prediction = tf.equal(tf.argmax(y_, axis = 1), tf.argmax(y, axis = 1))
#正确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction ,tf.float32))


for i in range(1000):
    #从数据集里面读取100张图片和标签
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict = {x:batch_xs, y:batch_ys})
    if(i % 100 == 0):
        print("损失值",sess.run(loss, feed_dict={x: mnist.test.images, y: mnist.test.labels}))
        print('识别正确率',sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels}))

#导入图片处理相关库
from PIL import Image
from itertools import chain

#输入一张图片，转换为张量
test_img = Image.open("./2.bmp")   #打开图片
test_img_array = np.array(test_img)#将图片数据存放到数组
test_img_array = np.asarray(test_img_array, dtype="float32")#转换数组元素的数据类型
#转换数组成张量
test_img_array = list(chain.from_iterable(test_img_array))#一阶
test_img_array = [test_img_array]#二阶

#还原计算图
layer1 = tf.nn.relu(tf.matmul(test_img_array, w1) + b1)
y_test = tf.nn.softmax(tf.matmul(layer1,w2) + b2)

#启动会话推理

#打印结果
print("识别结果：")
print(sess.run(y_test))

print("保存图片和参数：")
with open("img2.h", "w") as f:
    img_save = str(test_img_array)
    str1 = img_save.replace("[", "")
    str2 = str1.replace("]", "")
    f.write("img2[784]={"+str2+"};")

with open("w1.h", "w") as f:
    w1_value = w1.eval(session=sess)#获取张量的值
    str1 = str(np.transpose(w1_value).tolist())#转换为字符
    str2 = str1.replace("[", "")
    str3 = str2.replace("]", "")
    f.write("w1[784*64]={"+str3+"};")
    #f.write(str1)
    
print("保存完成")