摘要
为解决传统人工巡检效率低、检测精度不高的问题,提出了一种基于改进YOLOv8的小目标多类别粮仓害虫实时检测算法。采用RepGFPN结构替代传统的FPN-PANet架构,增强多尺度特征融合的能力。采用FocalNets替代YOLOv8中的SPPF模块,利用焦点调制机制提升对小目标细节和复杂背景的处理能力。结合CGA模块通过级联注意力机制减少跨层冗余信息,提升模型对关键目标的检测精度。在SGI-6数据集上对比改进YOLOv8和YOLOv5、YOLOv4和Faster R-CNN的检测性能,并开展消融实验。结果表明,相较于对比算法,改进YOLOv8在mAP、帧率和推理时间方面均表现出色,同时各个改进模块的逐步加入显著提升了模型的性能。该算法能够有效满足粮仓害虫检测的实际需求,为智能监控系统的应用提供了可行的解决方案。
To address the issues of low efficiency and insufficient detection accuracy in traditional manual inspections,a real-time detection algorithm for small target multi-class pests in grain warehouses based on an improved YOLOv8 is proposed.The RepGFPN structure is used to replace the traditional FPN-PANet architecture,enhancing the capability of multi-scale feature fusion.FocalNets is adopted to replace the SPPF module in YOLOv8,utilizing a focal modulation mechanism to improve the processing of small target details and complex backgrounds.The CGA module is integrated with a cascaded attention mechanism to reduce cross-layer redundant information,thereby enhancing the model's detection accuracy for key targets.The detection performance of the improved YOLOv8 is compared with YOLOv5,YOLOv4,and Faster R-CNN on the SGI-6 dataset,and ablation experiments are conducted.The results show that,compared to the baseline algorithms,the improved YOLOv8 performs excellently in terms of mAP,frame rate,and inference time.Additionally,the progressive integration of each improved module significantly enhances the model's performance.This proposed algorithm effectively meets the practical requirements of detection of pests in grain warehouse,providing a feasible solution for the application of intelligent monitoring systems.
作者
訾永所
洪银胜
江艳琼
ZI Yong-suo;HONG Yin-sheng;JIANG Yan-qiong(Kunming Metallurgy College,Kunming 650033,China;Yunnan College of Business Management,Kunming 650031,China)
出处
《粮食与饲料工业》
2026年第1期41-46,共6页
Cereal & Feed Industry
基金
全国高等院校计算机基础教育研究会研究项目(2025-AFCEC-741)
昆明冶金高等专科学校科研基金项目(2024xjz02)
教育部信息化教指委全国高等职业院校信息技术与人工智能通识课程教学改革研究项目(KT2502070)。
关键词
YOLOv8
实时检测
小目标
粮仓害虫
多类别
YOLOv8
real-time detection
small target
pests in grain warehouses
multi-class
