weak correlation features (like cats and cars in the
first data set), while when processing data with strong
correlation features (such as different models
between cars in the second data set), the deviation
between the test accuracy of each feature is small.
4 CONCLUSION
This research discusses the performance of VGG16
model in dealing data sets of different complexity and
kinds through two experiments. The results of
experiments show that VGG16 model has great
performance on the simpler CIFAR-10 with
exactness rates of 84.45%, however the performance
of the VGG16 model decreased when it deals with
complex BIT-Vehicle data set, and the training
accuracy dropped to 77.41%. The accuracy of all
categories in BIT-Vehicle data set are less than 80%.
The results of research show that the performance
of CNN may be influenced in dealing datasets with
different kinds and more span range. It may need
more training samples and deeper network structure
to extract effective features especially facing data sets
with high complexity. Furthermore, the recognition
accuracy of different categories exists big differences
in the BIT-Vehicle data set. This may be due to the
sample amounts of different categories in the data set
is imbalance, or the complexity of the different
categories is too high.
This research not only presents the challenges
CNN may face while dealing with complicated data
sets, but also provides a framework for evaluation and
comparison of CNN. It provides a significant
reference for subsequent image recognition in
practical applications.
In general, this research provides a preliminary
framework for evaluating and comparing the
performance of CNN when processing data sets with
different features. In future research, the exploration
of optimizing the parameters and structure of the
CNN model to improve its performance on different
data sets will be promoted further. Moreover, it is
necessary to design a more efficient model which can
fit with further and deeper data or use more data to
train and test while facing with diverse type of
features.
AUTHORS CONTRIBUTION
All the authors contributed equally and their names
were listed in alphabetical order.
REFERENCES
Aytekin, A., Mençik, V., 2022. Detection of Driver
Dynamics with VGG16 Model. Applied Computer
Systems, 27(1): 83-88.
Bensedik, H., Azough, A., Meknasssi, M., 2018. Vehicle
type classification using convolutional neural network,
In 2018 IEEE 5th International Congress on
Information Science and Technology (CiSt). Institute of
Electrical and Electronic Engineers, 313-316.
Chauhan, M.S., Singh, A., Khemka, M. et al., 2019.
Embedded CNN Based Vehicle Classification and
Counting in Non-Laned Road Traffic. Information and
Communication Technologies and Development, 1-11.
Chen, Y., Zhu, W., Yao, D., Zhang, L., 2017. Vehicle Type
Classification based on Convolutional Neural Network.
In 2017 Chinese Automation Congress (CAC). Institute
of Electrical and Electronic Engineers, 1898-1901.
Dhuri, V., Khan, A., Kamtekar, Y. et al., 2021. Real-Time
Parking Lot Occupancy Detection System with VGG16
Deep Neural Network using Decentralized Processing
for Public, Private Parking Facilities. In 2021
International Conference on Advances in Electrical,
Computing, Communication and Sustainable
Technologies (ICAECT), 1-8.
Gao, Y., Xiao, G., 2021. Real-time Chinese Traffic
Warning Signs Recognition Based on Cascade and
CNN. Journal of Real-Time Image Processing, 18(3):
669–680.
Li, S., Lin, J., Li, G. et al., 2018. Vehicle Type Detection
Based on Deep Learning in Traffic Scene. Procedia
Computer Science, 131: 564-572.
Li, Y., Song, B., Kang, X. et al., 2018. Vehicle-Type
Detection Based on Compressed Sensing and Deep
Learning in Vehicular Networks. Sensors, 18(12): 4500.
Liu, H., Wang, H., 2023. Real-Time Anomaly Detection of
Network Traffic Based on CNN. Symmetry, 15(6).
Njayou, Y., 2022. Traffic Sign Classification Using CNN
and Detection Using Faster-RCNN and YOLOV4.
Heliyon, 8(12).
Park, S.H., Yu, S.B., Kim, J.A. et al., 2022. An All-in-One
Vehicle Type and License Plate Recognition System
Using YOLOv4. Sensors, 22(3): 921.
Qiao, X., 2023. Research on Traffic Sign Recognition based
on CNN Deep Learning Network. Procedia Computer
Science, 228: 826-837.
Sang, J., Wu, Z., Guo, P. et al., 2018. An Improved
YOLOv2 for Vehicle Detection. Sensors, 18(12): 4272.