Tensorflow VGG网络构建手写数字识别系统设计--代码篇

Tensorflow VGG网络构建手写数字识别系统设计

 

 

1、建立VGG16模型

class VGG16_Mode():
    """ create vgg16 network use tensorflow
        VGG16 network structure:
        (conv 3x3 64)=>(conv 3x3 64, pool/2)
        (conv 3x3 128)=>(conv 3x3 128, pool/2)
        (conv 3x3 256)=>(conv 3x3 256)=>(conv 3x3 256)=>(conv 3x3 256, pool/2)
        (conv 3x3 512)=>(conv 3x3 512)=>(conv 3x3 512)=>(conv 3x3 512, pool/2)
        (fc 4096)=>(fc 4096)=>(fc classes)
    """
    
    def variable_summaries(self, var, name):
        with tf.name_scope('summaries'):
            tf.summary.histogram(name, var)
            mean = tf.reduce_mean(var)
            tf.summary.scalar('mean/' + name, mean)
            stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
            tf.summary.scalar('stddev/' + name, stddev)  
            
            
    def conv_layer(self, data, ksize, stride, name, w_biases = False,padding = "SAME"):
        with tf.variable_scope(name) as scope:
            with tf.name_scope(name):
                with tf.name_scope("weights"):
                    w_init = tf.contrib.layers.xavier_initializer()
                    w = tf.get_variable(shape= ksize, initializer= w_init,name= 'w')
                    self.variable_summaries(w, name + "/weights")
                with tf.name_scope("biases"):
                    biases = tf.Variable(tf.constant(0.0, shape=[ksize[3]], dtype=tf.float32), 'biases')
                    self.variable_summaries(biases, name + "/biases")
                with tf.name_scope("Wx_plus_b"):
                    if w_biases == False:
                        cov = tf.nn.conv2d(input= data, filter= w, strides= stride, padding= padding)
                    else:
                        cov = tf.nn.conv2d(input= data,filter= w, stride= stride,padding= padding) + biases       
        
        # 记录神经网络节点输出在经过激活函数之后的分布。
        tf.summary.histogram(name + '/activations', cov)
        return cov

    def pool_layer(self, data, ksize, stride, name, padding= 'VALID'):
        with tf.variable_scope(name) as scope:
            max_pool =  tf.nn.max_pool(value= data, ksize= ksize, strides= stride,padding= padding)
        return max_pool

    def flatten(self,data):
        [a,b,c,d] = data.shape
        ft = tf.reshape(data,[-1,b*c*d])
        return ft

    def fc_layer(self,data,name,fc_dims):
        with tf.variable_scope(name) as scope:
            with tf.name_scope(name):
                with tf.name_scope("weights"):
                    w_init = tf.contrib.layers.xavier_initializer()
                    w = tf.get_variable(shape=[data.shape[1],fc_dims],name= 'w',initializer=w_init)
                    self.variable_summaries(w, name + "/weights")
                with tf.name_scope("biases"):
                    biases = tf.Variable(tf.constant(0.0, shape=[fc_dims], dtype=tf.float32), 'biases')
                    self.variable_summaries(biases, name + "/biases")
                with tf.name_scope("Wx_plus_b"):
                    fc = tf.nn.relu(tf.matmul(data,w)+ biases)
                    tf.summary.histogram(name + '/pre_activations', tf.matmul(data,w)+ biases)       
        
        # 记录神经网络节点输出在经过激活函数之后的分布。
        tf.summary.histogram(name + '/activations', fc)
        return fc

    def finlaout_layer(self,data,name,fc_dims):
        with tf.variable_scope(name, reuse=tf.AUTO_REUSE):
            with tf.name_scope(