面向复杂工业场景的计算机视觉轻量化检测方法研究

Research on lightweight detection methods for computer vision in complex industrial scenarios

摘要

复杂工业场景存在光照多变、背景杂乱、目标微小等问题，传统计算机视觉检测模型难以兼顾检测精度与轻量化部署需求。本文梳理基于网络架构优化、模型压缩、迁移与小样本学习、多模态融合的四类轻量化检测方法，对比各类方法优劣，结合应用实践分析关键问题与解决方案，通过构建评估指标体系和对比实验验证方法有效性。研究表明，融合多技术的轻量化方法能更好适配复杂工业场景，实现精度与效率的平衡，为工业检测智能化、轻量化落地提供理论支撑。

Abstract

Complex industrial scenes present challenges such as variable illumination, cluttered backgrounds, and small targets, making it difficult for traditional computer vision detection models to balance detection accuracy and lightweight deployment requirements. This paper reviews four types of lightweight detection methods based on network architecture optimization, model compression, transfer and few-shot learning, and multi-modal fusion. It compares the advantages and disadvantages of various methods, analyzes key issues and solutions in combination with application practices, and verifies the effectiveness of the methods through the construction of an evaluation index system and comparative experiments. The research shows that lightweight methods integrating multiple technologies can better adapt to complex industrial scenes, achieving a balance between accuracy and efficiency, and providing theoretical support for the implementation of intelligent and lightweight industrial detection.

关键词

复杂工业场景 / 计算机视觉 / 轻量化检测 / 多模态融合

Key words

Complex industrial scene / Computer vision / Lightweight detection / Multimodal fusion

参考文献

[1] 杨育学, 林炜阳, 王博, 等 . 基于计算机视觉的PCB异常检测系统[J]. 物联网技术, 2025, 15(21): 43-49.

[2] 张琦. 基于计算机视觉的深度学习图像识别算法优化研究[J]. 信息与电脑, 2025, 37(19): 1-3.

[3] 孙仁科, 营鹏, 李仲年, 等 . 基于轻量化SSD的弱小目标检测[J]. 计算机仿真, 2024, 41(10): 355-361.