关键词:
多目标检测
YOLOv8
轻量化
共享卷积
Wise-IoU
摘要:
本文提出了一种多机制融合的轻量化交通多目标检测算法,该算法以YOLOv8算法的基本网络结构为基础,通过引入多种机制对YOLOv8算法进行改进,以实现轻量化的目的,并且在确保轻量化的前提下,能够尽量保持YOLOv8算法的高检测精度。本文主要做了三方面的创新工作:1) 设计了FasterC2f模块,替代了YOLOv8算法中的C2f模块,该模块通过降维策略,对输入特征图进行高效处理,显著降低了模型的参数总量与计算复杂度;2) 设计了一种轻量化的检测头架构——Lightweight Shared Convolutional Detection (LSCD),通过引入共享卷积机制,有效减少了检测头的参数量,同时增强了特征图之间的全局信息融合能力,确保了较高的检测精度;3) 对损失函数进行改进,引入Wise-IoU机制,在不增加复杂度的前提下,进一步提高算法的检测精度。实验结果表明,本文提出的算法在交通多目标检测任务上取得了优异的表现,不仅在模型体积和计算量上实现了显著的轻量化,而且保持了较高的检测精度。This paper presents a lightweight multi-target traffic detection algorithm that integrates multiple mechanisms. Based on the fundamental network architecture of the YOLOv8 algorithm, this algorithm introduces various mechanisms to improve YOLOv8, aiming to achieve lightweightness while maintaining its high detection accuracy as much as possible. The main innovations of this paper lie in three aspects: 1) A FasterC2f module is designed to replace the C2f module in YOLOv8. This module employs a dimensionality reduction strategy to efficiently process input feature maps, significantly reducing the total number of model parameters and computational complexity. 2) A lightweight detection head architecture, named Lightweight Shared Convolutional Detection (LSCD), is devised. By incorporating a shared convolutional mechanism, it effectively decreases the number of parameters in the detection head while enhancing the global information fusion capability among feature maps, thereby ensuring high detection accuracy. 3) The loss function is improved by introducing the Wise-IoU mechanism, which further enhances the detection accuracy of the algorithm without increasing its complexity. Experimental results demonstrate that the proposed algorithm exhibits excellent performance in multi-target traffic detection tasks, achieving remarkable lightweightness in terms of model size and computational cost while maintaining high detection accuracy.