Evolutionary neural architecture search for traffic sign recognition

Abstract

Convolutional neural networks (CNNs) exhibit superior performance in image feature extraction, making them extensively used in the area of traffic sign recognition. However, the design of existing traffic sign recognition algorithms often relies on expert knowledge to enhance the image feature extraction networks, necessitating image preprocessing and model parameter tuning. This increases the complexity of the model design process. This study introduces an evolutionary neural architecture search (ENAS) algorithm for the automatic design of neural network models tailored for traffic sign recognition. By integrating the construction parameters of residual network (ResNet) into evolutionary algorithms (EAs), we automatically generate lightweight networks for traffic sign recognition, utilizing blocks as the fundamental building units. Experimental evaluations on the German traffic sign recognition benchmark (GTSRB) dataset reveal that the algorithm attains a recognition accuracy of 99.32%, with a mere 2.8×10⁶ parameters. Experimental results comparing the proposed method with other traffic sign recognition algorithms demonstrate that the method can more efficiently discover neural network architectures, significantly reducing the number of network parameters while maintaining recognition accuracy.

References

[1] LiL F, WangY R. Improved LeNet-5 convolutional neural network traffic sign recognition[J]. International core journal of engineering, 2021, 7(4): 114-121

[2] ChengP, LiuW, ZhangY F, et al.. LOCO: local context based faster R-CNN for small traffic sign detection[C]. MultiMedia Modeling: 24th International Conference, February 5–7, 2018, Bangkok, Thailand, 2018 Heidelberg Springer International Publishing 329-341

[3] WeiT C, ChenX F, YinY L. Research on traffic sign recognition method based on multi-scale convolution neural network[J]. Journal of Northwestern Polytechnical University, 2021, 39(4): 891-899 in Chinese)

[4] MaY J, ChengS S, MaY T, et al.. Traffic sign recognition based on multi-scale feature fusion and extreme learning machine[J]. Chinese journal of liquid crystals and displays, 2020, 35(6): 572-582 (in Chinese)

[5] XueZ X, ZhengY H, XiaoJ, et al.. Traffic sign recognition based on multi-scale convolutional neural network[J]. Computer engineering, 2020, 46(3): 261-266 (in Chinese)

[6] HuangL, SunS Q, ZengJ, et al.. U-DARTS: uniform-space differentiable architecture search[J]. Information sciences, 2023, 628: 339-349

[7] YeP, LiB P, LiY K, et al.. B-darts: beta-decay regularization for differentiable architecture search[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 19–23, 2022, New Orleans, Louisiana, USA, 2022 New York IEEE 10874-10883

[8] StanleyK O, CluneJ, LehmanJ, et al.. Designing neural networks through neuroevolution[J]. Nature machine intelligence, 2019, 1(1): 24-35

[9] StanleyK O, MiikkulainenR. Evolving neural networks through augmenting topologies[J]. Evolutionary computation, 2002, 10(2): 99-127

[10] SongC W, MaY J, XuY, et al.. Multi-population evolutionary neural architecture search with stacked generalization[J]. Neurocomputing, 2024, 587: 127664

[11] XieY R, ChenH, MaY J, et al.. Automated design of CNN architecture based on efficient evolutionary search[J]. Neurocomputing, 2022, 491: 160-171

[12] GONG T, MA Y J, XU Y, et al. Efficient evolutionary neural architecture search based on hybrid search space[J]. International journal of machine learning and cybernetics, 2024: 1–14.

[13] HUANG J H, XUE B, SUN Y N, et al. Split-level evolutionary neural architecture search with elite weight inheritance[J]. IEEE transactions on neural networks and learning systems, 2023.

[14] XueY, ChenC, SłowikA. Neural architecture search based on a multi-objective evolutionary algorithm with probability stack[J]. IEEE transactions on evolutionary computation, 2023, 27(4): 778-786

[15] XUE Y, TONG W N, NERI F, et al. Evolutionary architecture search for generative adversarial networks based on weight sharing[J]. IEEE transactions on evolutionary computation, 2023.

[16] MaY J, LiuP P. The evolutionary design of convolutional neural network based on DenseNet for image classification[J]. Laser & optoelectronics progress, 2020, 57(24): 42-49 (in Chinese)

[17] CireanD, MeierU, MasciJ, et al.. Multi-column deep neural network for traffic sign classification[J]. Neural networks, 2012, 32: 333-338

[18] WangX B, HuangJ J, LiuW J. Traffic sign recognition based on optimized convolutional neural network architecture[J]. Journal of computer applications, 2017, 37(2): 530-534

[19] SongQ S, ZhangC, TianZ X, et al.. Traffic sign recognition based on multi-scale convolutional neural network[J]. Journal of Human University (Natural Sciences), 2018, 045(008): 131-137 (in Chinese)

[20] SunW, DuH J, ZhangX R, et al.. Traffic sign recognition method based on multi-layer feature CNN and extreme learning machine[J]. Journal of University of Electronic Science and Technology of China, 2018, 47(003): 343-349 (in Chinese)

RIGHTS & PERMISSIONS

Tianjin University of Technology