predict.py

import tensorflow as tf
import numpy as np
import os,glob,cv2
import sys,argparse


# 预测部分的初始化要跟模型部分的初始化一致
image_size=64
num_channels=3
images = []

# 待测试的图片，放到和程序同一目录下
path = 'island1.jpg'
image = cv2.imread(path)

image = cv2.resize(image, (image_size, image_size),0,0, cv2.INTER_LINEAR)
images.append(image)
images = np.array(images, dtype=np.uint8)
images = images.astype('float32')
images = np.multiply(images, 1.0/255.0) 

# tf要求输入是4维的：第一个参数是1因为每次预测的就是一张图  1 64 64 3
x_batch = images.reshape(1, image_size,image_size,num_channels)

# 开启一个对话
sess = tf.Session()
# 拿到刚才保存的模型：直接输文件名，要.meta结尾的
saver = tf.train.import_meta_graph('./model/island-seaice.ckpt-3960.meta')
# 拿刚才保存模型的权重参数：直接输文件名，后缀最长的那个
saver.restore(sess, './model/island-seaice.ckpt-3960')

# 同样导入标签：1和0
graph = tf.get_default_graph()

y_pred = graph.get_tensor_by_name("y_pred:0")

x= graph.get_tensor_by_name("x:0") 
y_true = graph.get_tensor_by_name("y_true:0") 
y_test_images = np.zeros((1, 2)) 

# 两个值：谁的概率大就是谁
feed_dict_testing = {x: x_batch, y_true: y_test_images}
# 预测开始：
result=sess.run(y_pred, feed_dict=feed_dict_testing)

# 把1和0又回到其对应的种类
res_label = ['island','seaice']
print('预测结果为:',res_label[result.argmax()])