基于Vgg-Unet模型自动驾驶场景检测

1.VGG

VGG全称是Visual Geometry Group属于牛津大学科学工程系，其发布了一些列以VGG开头的卷积网络模型，可以应用在人脸识别、图像分类等方面,VGG的输入被设置为大小为224x244的RGB图像。为训练集图像上的所有图像计算平均RGB值，然后将该图像作为输入输入到VGG卷积网络。使用3x3或1x1滤波器，并且卷积步骤是固定的。有3个VGG全连接层，根据卷积层+全连接层的总数，可以从VGG11到VGG19变化。最小VGG11具有8个卷积层和3个完全连接层。最大VGG19具有16个卷积层+3个完全连接的层。此外，VGG网络后面没有每个卷积层后面的池化层，也没有分布在不同卷积层下的总共5个池化层。
结构图如下：

架构图

2.Unet模型：

Unet是一个优秀的语义分割模型，其主要执行过程与其它语义分割模型类似。与CNN不同的之处在于CNN是图像级的分类，而unet是像素级的分类，其输出的是每个像素点的类别

主要代码如下：

def get_vgg_encoder(input_height=224,  input_width=224, channels=3):

    if channel == 'channels_first':
        img_input = Input(shape=(channels, input_height, input_width))
    elif channel == 'channels_last':
        img_input = Input(shape=(input_height, input_width, channels))

    x = Conv2D(64, (3, 3), activation='relu', padding='same',
               name='block1_conv1', data_format=channel)(img_input)
    x = Conv2D(64, (3, 3), activation='relu', padding='same',
               name='block1_conv2', data_format=channel)(x)
    x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool',
                     data_format=channel)(x)
    f1 = x
    # Block 2
    x = Conv2D(128, (3, 3), activation='relu', padding='same',
               name='block2_conv1', data_format=channel)(x)
    x = Conv2D(128, (3, 3), activation='relu', padding='same',
               name='block2_conv2', data_format=channel)(x)
    x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool',
                     data_format=channel)(x)
    f2 = x

    # Block 3
    x = Conv2D(256, (3, 3), activation='relu', padding='same',
               name='block3_conv1', data_format=channel)(x)
    x = Conv2D(256, (3, 3), activation='relu', padding='same',
               name='block3_conv2', data_format=channel)(x)
    x = Conv2D(256, (3, 3), activation='relu', padding='same',
               name='block3_conv3', data_format=channel)(x)
    x = MaxPooling2D((2, 2), strides=(2, 2), name='block3_pool',
                     data_format=channel)(x)
    f3 = x

    # Block 4
    x = Conv2D(512, (3, 3), activation='relu', padding='same',
               name='block4_conv1', data_format=channel)(x)
    x = Conv2D(512, (3, 3), activation='relu', padding='same',
               name='block4_conv2', data_format=channel)(x)
    x = Conv2D(512, (3, 3), activation='relu', padding='same',
               name='block4_conv3', data_format=channel)(x)
    x = MaxPooling2D((2, 2), strides=(2, 2), name='block4_pool',
                     data_format=channel)(x)
    f4 = x

    # Block 5
    x = Conv2D(512, (3, 3), activation='relu', padding='same',
               name='block5_conv1', data_format=channel)(x)
    x = Conv2D(512, (3, 3), activation='relu', padding='same',
               name='block5_conv2', data_format=channel)(x)
    x = Conv2D(512, (3, 3), activation='relu', padding='same',
               name='block5_conv3', data_format=channel)(x)
    x = MaxPooling2D((2, 2), strides=(2, 2), name='block5_pool',
                     data_format=channel)(x)
    f5 = x

    return img_input, [f1, f2, f3, f4, f5]

def _unet(classes, encoder, l1_skip_conn=True, input_height=416,
          input_width=608, channels=3):

    img_input, levels = encoder(
        input_height=input_height, input_width=input_width, channels=channels)
    [f1, f2, f3, f4, f5] = levels

    o = f4

    o = (ZeroPadding2D((1, 1), data_format=channel))(o)
    o = (Conv2D(512, (3, 3), padding='valid' , activation='relu' , data_format=channel))(o)
    o = (BatchNormalization())(o)

    o = (UpSampling2D((2, 2), data_format=channel))(o)
    o = (concatenate([o, f3], axis=-1))
    o = (ZeroPadding2D((1, 1), data_format=channel))(o)
    o = (Conv2D(256, (3, 3), padding='valid', activation='relu' , data_format=channel))(o)
    o = (BatchNormalization())(o)

    o = (UpSampling2D((2, 2), data_format=channel))(o)
    o = (concatenate([o, f2], axis=-1))
    o = (ZeroPadding2D((1, 1), data_format=channel))(o)
    o = (Conv2D(128, (3, 3), padding='valid' , activation='relu' , data_format=channel))(o)
    o = (BatchNormalization())(o)

    o = (UpSampling2D((2, 2), data_format=channel))(o)

    if l1_skip_conn:
        o = (concatenate([o, f1], axis=-1))

    o = (ZeroPadding2D((1, 1), data_format=channel))(o)
    o = (Conv2D(64, (3, 3), padding='valid', activation='relu', data_format=channel, name="seg_feats"))(o)
    o = (BatchNormalization())(o)

    o = Conv2D(classes, (3, 3), padding='same',
               data_format=channel)(o)

    model = get_segmentation_model(img_input, o)

    return model
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98