Research on Lightweight Convolutional Neural Network Algorithm Deployment and Performance Enhancement for Embedded Devices

Li Jinchao(Xijing University)
Li Sheng(Xijing University)

Abstract

In the context of the rapid development of intelligent terminals and the Internet of Things (IoT), Convolutional Neural Networks (CNNs) have achieved remarkable results in fields such as image recognition, object detection, and semantic segmentation. However, traditional deep networks are characterized by large structures, complex computations, and high energy consumption, making them difficult to deploy directly on resource-constrained platforms such as embedded devices. This paper systematically analyzes the core principles, optimization strategies, and deployment methods of lightweight CNNs, focusing on techniques such as model compression, network pruning, parameter quantization, and knowledge distillation. Combined with the concept of software–hardware co-optimization, it explores efficient mapping and performance improvement paths of models on embedded platforms. Through experimental analysis of representative lightweight networks (e.g., MobileNet, ShuffleNet, GhostNet), it is verified that model parameters and computation can be significantly reduced while maintaining accuracy. The study shows that the co-optimization of hardware characteristics and algorithm structures is the key direction for achieving high performance and low power consumption balance in intelligent inference on embedded devices.

Keywords

lightweight convolutional neural network; embedded device; model compression; parameter quantization; performance optimization

References

Xiao J ,Xu L ,Li C , et al.Lightweight visible damage detection algorithm for embedded systems applied to pipeline automation equipment[J].Journal of Pipeline Science and Engineering,2025,5(2):100254-100254.

Anver R S ,Deepambika A V ,Rahiman A M , et al.Emotional Speech Generation: An Approach Using Convolutional Neural Networks (CNN) Based Generative Adversarial Network[J].Circuits, Systems, and Signal Processing,2025,44(11):1-23.

Murti A M ,Setianingsih C ,Kusumawardhani E , et al.Cedarwood Quality Classification using SVM Classifier and Convolutional Neural Network (CNN)[J].International Journal of Advanced Computer Science and Applications (IJACSA),2022,13(11):

Kavita B ,Vijaya M .Evaluation of CNN model by comparing with convolutional autoencoder and deep neural network for crop classification on hyperspectral imagery[J].Geocarto International,2020,37(3):1-15.

Mousavi S M ,Rahmani F .Improving the detection efficiency of IRan ANtineutrino Detector (IRAND) based on Convolutional Neural Network (CNN)[J].Nuclear Engineering and Technology,2026,58(5):104118-104118.

Jinchao Li and Sheng Li are with the School of Electronic Information Engineering, Xijing University, Xi’an 710123, China (e-mail:[e-mail];[e-mail]

DOI: http://dx.doi.org/10.26549/met.v9i1.36160