Automatic Traffic State Recognition Based on Video Features Extracted by an Autoencoder

Table 6

Test results of the CNN models and the models proposed in this paper.

Model	Model structure	Accuracy rate (%)	Recall rate (%)	(%)
AlexNet	5 convolutional layers, 3 pooling layers, 3 fully connected layers, and 1 classification layer	94.5	93.6	94.0
LeNet	3 convolutional layers, 2 pooling layers, 1 fully connected layer, and 1 classification layer	82.3	62.4	71.0
GoogLeNet	22 network layers, including convolutional layers and pooling layers, and 1 classification layer	36.8	35.2	36.0
VGG16	13 convolutional layers, 3 fully connected layers, 5 pooling layers, and 1 classification layer	11.1	33.3	16.7
Classifier	5 encoding hidden layers + common classifiers (linear classification, SVM, DNN, and so on)	94.5–97.1	94.5–97.1	94.4–97.1
-k-means	5 encoding hidden layers + k-means clustering	95.4	95.3	95.3