tensorflow2.0手势识别【图片分类练手】

写在前面：感谢阿里天池实验室，让我免费白嫖到GPU训练数据

本文参考价值不大，只是本人练手的代码，简单保存下代码而已，若有错误，尽管提出，不胜感激。
-------------------------------------分割线--------------------------------

1-导入所需要的包

import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import os
import re
from PIL import Image
import random
%matplotlib inline

2-从文件中获取数据集

def make_data():   
    file_list = os.listdir('dataset/')
    pic_to_label = {'nothing':0,'ok':1,'peace':2,'punch':3,'stop':4}
    data_path = []
    data = []
    label = []

    for item in file_list:
        data_path.append(item)
        
    for i in range(10):
        random.shuffle(data_path)
    
    for item in data_path:
        temp = 'dataset/'+item
        if re.match('n+othing',item) != None:
            label.append([0])
            im  = Image.open(temp)
            im = im.convert('L')
            img_ndarray = np.asarray(im, dtype='float64')
            data.append(img_ndarray)
        elif re.match('iiok',item) != None:
            label.append([1])
            im  = Image.open(temp)
            im = im.convert('L')
            img_ndarray = np.asarray(im, dtype='float64')
            data.append(img_ndarray)
        elif re.match('peace',item) != None:
            label.append([2])
            im  = Image.open(temp)
            im = im.convert('L')
            img_ndarray = np.asarray(im, dtype='float64')
            data.append(img_ndarray)
        elif re.match('punch',item) != None:
            label.append([3])
            im  = Image.open(temp)
            im = im.convert('L')
            img_ndarray = np.asarray(im, dtype='float64')
            data.append(img_ndarray)
        elif re.match('stop',item) != None:
            label.append([4])
            im  = Image.open(temp)
            im = im.convert('L')
            img_ndarray = np.asarray(im, dtype='float64')
            data.append(img_ndarray)
    return data,label

3-加载数据集，分为训练集和测试集（6:4）

def load_data():
    data,label = make_data()
    data_number = len(data)
    train_number = data_number*6//10
    test_data = data_number-train_number

    data_train = data[0:train_number]
    label_train = label[0:train_number]
    data_test = data[train_number:-1]
    label_test = label[train_number:-1]

    return(data_train,label_train),(data_test,label_test)

4-预处理数据（归一化，标签变one_hot）

def process_data():
    (data_train,label_train),(data_test,label_test) = load_data()

    data_train = np.array(data_train)
    #data_train =  data_train.reshape(data_train.shape[0],1,200,200)/255
    data_test = np.array(data_test)
   # data_test = data_test.reshape(data_test.shape[0],1,200,200)/255

    data_train = np.expand_dims(data_train,-1)/255
    data_test = np.expand_dims(data_test,-1)/255

    label_train = np.array(label_train,dtype='uint8')
    label_test = np.array(label_test,dtype='uint8')
    
    label_train_onehot = tf.keras.utils.to_categorical(label_train)
    label_test_onehot = tf.keras.utils.to_categorical(label_test)
    
    return(data_train,label_train_onehot),(data_test,label_test_onehot)

5-搭建卷积网络

def build_model():
    model = tf.keras.Sequential()
    model.add(tf.keras.layers.Conv2D(16,(3,3),input_shape=(200,200,1),activation='relu',padding='same'))
    model.add(tf.keras.layers.Conv2D(16,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.MaxPool2D())
    model.add(tf.keras.layers.Dropout(0.3))

    model.add(tf.keras.layers.Conv2D(32,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.Conv2D(32,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.MaxPool2D())
    model.add(tf.keras.layers.Dropout(0.3))
    
    model.add(tf.keras.layers.Conv2D(64,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.Conv2D(64,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.MaxPool2D())
    model.add(tf.keras.layers.Dropout(0.3))
    
    model.add(tf.keras.layers.Conv2D(128,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.Conv2D(128,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.MaxPool2D())
    model.add(tf.keras.layers.Dropout(0.3))

    model.add(tf.keras.layers.Conv2D(256,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.Conv2D(256,(3,3),activation='relu',padding='same'))
    model.add(tf.keras.layers.MaxPool2D())
    model.add(tf.keras.layers.Dropout(0.3))


    model.add(tf.keras.layers.GlobalAveragePooling2D())


    model.add(tf.keras.layers.Dense(256,activation='relu'))
    model.add(tf.keras.layers.Dropout(0.3))

    model.add(tf.keras.layers.Dense(256,activation='relu'))
    model.add(tf.keras.layers.Dropout(0.3))

    model.add(tf.keras.layers.Dense(5,activation='softmax'))
    
    return model

6-后续训练数据，预测数据

model = build_model()
model.summary()

model.compile(optimizer='adam',
              loss='categorical_crossentropy',
              metrics=['acc']
)

(data_train,label_train),(data_test,label_test) = process_data()

history = model.fit(data_train,label_train,epochs=30,validation_data=(data_test,label_test))

history.history.keys()     

plt.plot(history.epoch,history.history.get('loss'),label='loss')
plt.plot(history.epoch,history.history.get('val_loss'),label='val_loss')
plt.legend()

plt.plot(history.epoch,history.history.get('val_acc'),label='val_acc')
plt.plot(history.epoch,history.history.get('acc'),label='acc')
plt.legend()

model.evaluate(data_test,label_test)

7-训练过程情况

8-训练结果测试

pic_to_label = {0:'nothing',1:'ok',2:'peace',3:'punch',4:'stop'}
im = Image.open('dataset/punch100.png')
im = im.convert('L')
im = np.asarray(im, dtype='float64')
plt.imshow(im)
im = im.reshape(1,200,200,1)
ans = model.predict(im)
ans = np.argmax(ans)
print('                ',pic_to_label[ans])