PSO-based lightweight neural architecture search for object detection

IF 8.2 1区 计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE Swarm and Evolutionary Computation Pub Date : 2024-08-02 DOI:10.1016/j.swevo.2024.101684
Tao Gong, Yongjie Ma
{"title":"PSO-based lightweight neural architecture search for object detection","authors":"Tao Gong,&nbsp;Yongjie Ma","doi":"10.1016/j.swevo.2024.101684","DOIUrl":null,"url":null,"abstract":"<div><p>In recent years, Neural Architecture Search (NAS) has received widespread attention for its remarkable ability to design neural networks. However, existing NAS works have mainly focused on network search, with limited emphasis on downstream applications after discovering efficient neural networks. In this paper, we propose a lightweight search strategy based on the particle swarm optimization algorithm and apply the searched network as backbone for object detection tasks. Specifically, we design a lightweight search space based on Ghostconv modules and improved Mobileblocks, achieving comprehensive exploration within the search space using variable-length encoding strategy. During the search process, to balance network performance and resource consumption, we propose a multi-objective fitness function and incorporated the classification accuracy, parameter size, and FLOPs of candidate individuals into optimization. For particle performance evaluation, we propose a new strategy based on weight sharing and dynamic early stopping, significantly accelerating the search process. Finally, we fine-tune the globally optimal particle decoded as the backbone, adding Ghost PAN feature fusion modules and detection heads to build an object detection model, and we achieve a 17.01% mAP on the VisDrone2019 dataset. Experimental results demonstrate the competitiveness of our algorithm in terms of search time and the balance between accuracy and efficiency, and also confirm the effectiveness of object detection models designed through NAS methods.</p></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"90 ","pages":"Article 101684"},"PeriodicalIF":8.2000,"publicationDate":"2024-08-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Swarm and Evolutionary Computation","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2210650224002220","RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0

Abstract

In recent years, Neural Architecture Search (NAS) has received widespread attention for its remarkable ability to design neural networks. However, existing NAS works have mainly focused on network search, with limited emphasis on downstream applications after discovering efficient neural networks. In this paper, we propose a lightweight search strategy based on the particle swarm optimization algorithm and apply the searched network as backbone for object detection tasks. Specifically, we design a lightweight search space based on Ghostconv modules and improved Mobileblocks, achieving comprehensive exploration within the search space using variable-length encoding strategy. During the search process, to balance network performance and resource consumption, we propose a multi-objective fitness function and incorporated the classification accuracy, parameter size, and FLOPs of candidate individuals into optimization. For particle performance evaluation, we propose a new strategy based on weight sharing and dynamic early stopping, significantly accelerating the search process. Finally, we fine-tune the globally optimal particle decoded as the backbone, adding Ghost PAN feature fusion modules and detection heads to build an object detection model, and we achieve a 17.01% mAP on the VisDrone2019 dataset. Experimental results demonstrate the competitiveness of our algorithm in terms of search time and the balance between accuracy and efficiency, and also confirm the effectiveness of object detection models designed through NAS methods.

查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
基于 PSO 的轻量级目标检测神经架构搜索
近年来,神经架构搜索(NAS)因其卓越的神经网络设计能力而受到广泛关注。然而,现有的 NAS 作品主要集中于网络搜索,对发现高效神经网络后的下游应用重视有限。在本文中,我们提出了一种基于粒子群优化算法的轻量级搜索策略,并将搜索到的网络作为骨干应用于物体检测任务。具体来说,我们基于 Ghostconv 模块和改进的 Mobileblocks 设计了一个轻量级搜索空间,利用变长编码策略实现了搜索空间内的全面探索。在搜索过程中,为了平衡网络性能和资源消耗,我们提出了多目标拟合函数,并将候选个体的分类精度、参数大小和 FLOPs 纳入优化。在粒子性能评估方面,我们提出了一种基于权重共享和动态提前停止的新策略,大大加快了搜索过程。最后,我们以解码的全局最优粒子为骨干进行微调,加入 Ghost PAN 特征融合模块和检测头,构建了物体检测模型,并在 VisDrone2019 数据集上实现了 17.01% 的 mAP。实验结果证明了我们的算法在搜索时间上的竞争力以及准确性和效率之间的平衡,同时也证实了通过 NAS 方法设计的物体检测模型的有效性。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 去求助
来源期刊
Swarm and Evolutionary Computation
Swarm and Evolutionary Computation COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCEC-COMPUTER SCIENCE, THEORY & METHODS
CiteScore
16.00
自引率
12.00%
发文量
169
期刊介绍: Swarm and Evolutionary Computation is a pioneering peer-reviewed journal focused on the latest research and advancements in nature-inspired intelligent computation using swarm and evolutionary algorithms. It covers theoretical, experimental, and practical aspects of these paradigms and their hybrids, promoting interdisciplinary research. The journal prioritizes the publication of high-quality, original articles that push the boundaries of evolutionary computation and swarm intelligence. Additionally, it welcomes survey papers on current topics and novel applications. Topics of interest include but are not limited to: Genetic Algorithms, and Genetic Programming, Evolution Strategies, and Evolutionary Programming, Differential Evolution, Artificial Immune Systems, Particle Swarms, Ant Colony, Bacterial Foraging, Artificial Bees, Fireflies Algorithm, Harmony Search, Artificial Life, Digital Organisms, Estimation of Distribution Algorithms, Stochastic Diffusion Search, Quantum Computing, Nano Computing, Membrane Computing, Human-centric Computing, Hybridization of Algorithms, Memetic Computing, Autonomic Computing, Self-organizing systems, Combinatorial, Discrete, Binary, Constrained, Multi-objective, Multi-modal, Dynamic, and Large-scale Optimization.
期刊最新文献
Complexity, interpretability and robustness of GP-based feature engineering in remote sensing A Self-adaptive two stage iterative greedy algorithm based job scales for energy-efficient distributed permutation flowshop scheduling problem Two-stage bidirectional coevolutionary algorithm for constrained multi-objective optimization A survey on vehicle–drone cooperative delivery operations optimization: Models, methods, and future research directions Bayesian network structure learning based on discrete artificial jellyfish search: Leveraging scoring and graphical properties
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1