A scene recognition algorithm based on deep residual network

Figures & data

Figure 1. Structure of a typical convolutional neural network.

Table 1. Comparison of convolutional neural network models.

Network model	Network depth	Additional technology
AlexNet (Krizhevsky, Sutskever, & Hinton, 2012)	5	–
ZF-Net (Zeiler & Fergus, 2014)	5	Network width increased
VGG_CNN_M/S/F (Chatfield et al., 2014)	5	Network width increased
VGG (Simonyan & Zisserman, 2014)	16–19	3 × 3 small convolution kernel parameter reduction
GoogLeNet/Inception V1 (Szegedy, Liu, et al., 2015)	22	Inception module, auxiliary classifier
Inception V2/V3 (Szegedy, Vanhoucke, Ioffe, Shlens, & Wojna, 2015)	46	Inception module improved Convolution kernel decomposition decompositionkernel decomposition
Residual-Net (He, Zhang, Ren, & Sun, 2015)	50/101/152	Residual connection
Inception-ResNet/Inception V4 (Szegedy, Ioffe, & Vanhoucke, 2016)	175	Combined with residual connection and inception module

Figure 2. Structure of scene image recognition model Scene-RecNet.

Figure 3. Examples of images in self-made data sets.

Table 2. Comparison of feature extraction performance.

Feature extraction algorithm	Feature dimension	Classification accuracy (%)
HSV	256	29.5
SIFT-PCA + BOW (1000)	1000	32.0
ScSPM (500)+PCA	1024	36.5
HOG + PCA	128	35.0
GIST	960	41.8
VGGNet-11	4096	86.9
VGGNet-16	4096	89.0
VGGNet-19	4096	88.1
GoogLeNet (Pool5)	1024	91.8
Scene-RecNet	512	94.4

Table 3. Comparison of network versatility.

Network model	With/without feature adjustment layer	Feature dimension	Classification accuracy (%)
VGGNet-19	×	4096	66.2
VGGNet-19	√	512	74.8
GoogLeNet	×	1024	70.0
GoogLeNet	√	512	76.8
ResNet	×	1024	74.6
Scene-RecNet	√	512	76.6

Figure 4. Comparison of feature dimension reduction effect.

Figure 5. Comparison of network depth increase effect.

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097–1105).

 Google Scholar

Zeiler, M. D., & Fergus, R. (2014). Visualizing and understanding convolutional networks. In European conference on computer vision (pp. 818–833).

 Google Scholar

Chatfield, K., Simonyan, K., Vedaldi, A., & Zisserman, A. (2014). Return of the devil in the details: Delving deep into convolutional nets. Computer Science, arXiv:1405.3531 v4 [cs.CV].

 Google Scholar

Simonyan, Karen, & Zisserman, Andrew. (2014). Very deep convolutional networks for large-scale image recognition. Computer Science, 1–14. arXiv:1409.1556 [cs.CV].

 Google Scholar

Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., … Rabinovich, A. (2015). Going deeper with convolutions. In IEEE conference on computer vision and pattern recognition (pp. 1–9).

 Google Scholar

Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., & Wojna, Z. (2015). Rethinking the inception architecture for computer vision. Retrieved from https://arxiv.org/pdf/1512.00567.pdf

 Google Scholar

He, K., Zhang, X., Ren, S., & Sun, J. (2015). Deep residual learning for image recognition. IEEE conference on computer vision and pattern recognition.

 Google Scholar

Szegedy, C., Ioffe, S., & Vanhoucke, V. (2016). Inception-v4, Inception-ResNet and the impact of residual connections on learning. Retrieved from https://arxiv.org/pdf/1602.07261.pdf

 Google Scholar