【DL项目】基于卷积神经网络实现人脸表情识别

• 更新时间：2018-07-05

前言

项目是好久之前做的，现在可能类似的demo也有很多了，已知忘记发博客和上传GitHub了，刚传上去，地址：https://github.com/roguesir/DL-ML-project/tree/master/Expression-Recognition

结构设计

采用fer2013人脸表情数据集，使用裁剪和旋转变换进行数据增强，处理后的数据量达到105000张，大小为48x48的黑白图片。

基本结构如下：

训练结果

模型	准确率（%）
model-basic	90.11
model-8	99.27
model-13	99.80
model-13-conv	99.86

部分代码

# -*- coding: utf-8 -*-

from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Flatten, Activation
from keras.layers.convolutional import Conv2D, MaxPooling2D
from keras.preprocessing.image import load_img, img_to_array
from keras.utils import np_utils
import os
import numpy as np


def load_dataset(filedir):
    """
    读取数据
    :param filedir:
    :return:
    """
    image_data_list = []
    label = []
    train_image_list = os.listdir(filedir + '/train_data')
    for img in train_image_list:
        url = os.path.join(filedir + '/train_data/' + img)
        image = load_img(url, grayscale=True, target_size=(48, 48))
        #print(image.shape)
        image_data_list.append(img_to_array(image))
        label.append(img.split('.')[0].split("-")[1])
    img_data = np.array(image_data_list)
    img_data = img_data.astype('float32')
    img_data /= 255
    return img_data, label


def make_network():
    model = Sequential()
    model.add(Conv2D(32, (3, 3), padding='same', input_shape=(48, 48, 1)))
    model.add(Activation('relu'))
    model.add(Conv2D(32, (3, 3)))
    model.add(Activation('relu'))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Dropout(0.5))

    model.add(Flatten())
    model.add(Dense(128))
    model.add(Activation('relu'))
    model.add(Dropout(0.5))
    model.add(Dense(7))
    model.add(Activation('softmax'))

    model.save('./data/model/model-basic.h5')
    return model


if __name__ == '__main__':
    train_loss = []
    train_accuracy = []
    train_x, train_y = load_dataset('data')
    train_y = np_utils.to_categorical(train_y)
    model = make_network()
    model.compile(loss='categorical_crossentropy', optimizer='adadelta', metrics=['accuracy'])
    model.fit(train_x, train_y, batch_size=500, epochs=200, verbose=1)