[1]王杰,马纪颖.融合改进的YOLOv5n和通道剪枝的寄生卵检测和分类[J].计算机技术与发展,2025,(02):146-152.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0295]
WANG Jie,MA Ji-ying.Detection and Classification of Parasitic Eggs Using a Fused Improved YOLOv5n and Channel Pruning Algorithm[J].,2025,(02):146-152.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0295]
点击复制
融合改进的YOLOv5n和通道剪枝的寄生卵检测和分类(
)
《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]
- 卷:
-
- 期数:
-
2025年02期
- 页码:
-
146-152
- 栏目:
-
人工智能
- 出版日期:
-
2025-02-10
文章信息/Info
- Title:
-
Detection and Classification of Parasitic Eggs Using a Fused Improved YOLOv5n and Channel Pruning Algorithm
- 文章编号:
-
1673-629X(2025)02-0146-07
- 作者:
-
王杰1; 2; 马纪颖1; 2
-
1. 沈阳化工大学 计算机科学与技术学院,辽宁 沈阳 110142;
2. 辽宁省化工过程工业智能化技术重点实验室,辽宁 沈阳 110142
- Author(s):
-
WANG Jie1; 2; MA Ji-ying1; 2
-
1. School of Computer Science and Technology,Shenyang University of Chemical Technology,Shenyang 110142,China;
2. Liaoning Key Laboratory of Intelligent Technology for Chemical Process Industry,Shenyang 110142,China
-
- 关键词:
-
YOLOv5算法; 寄生卵检测; C3_Faster; RepConv; 通道剪枝
- Keywords:
-
YOLOv5 algorithm; parasitic egg detection; C3_Faster; RepConv; channel pruning
- 分类号:
-
TP391.41
- DOI:
-
10.20165/j.cnki.ISSN1673-629X.2024.0295
- 摘要:
-
当前众多目标检测模型过于复杂,难以实现将寄生卵的检测和分类任务部署在移动设备,就此该文研究探讨了一种融合改进的 YOLOv5n 和通道剪枝的算法。 选择 YOLOv5 是由于 YOLOv5 的轻量化以及较高的精确度,能够达到该文的实验目的。 该文采用融合 C3_Faster 模块和 RepConv 重参数化模块对 YOLOv5n 的 BackBone 中的所有 C3 模块和 Neck 网络中部分卷积模块进行替换,C3_Faster 模块通过 PConv 减少卷积操作加快网络模型推理速度,RepConv 重参数化模块在训练阶段实行多分支结构增强特征提取能力,在验证阶段实行单分支结构加快检测速度,同时在改进后的 YOLOv5n 模型上进行稀疏训练和通道剪枝,通过减少模型中的冗余通道来降低模型复杂度、减少参数数量、提高检测效率和降低模型权重。 在寄生卵检测和分类任务对比实验中,该方法与 YOLOv5n、YOLOv5s、YOLOv7-tiny、YOLOv8n 和 SSD 目标检测算法相比,在检测精度略微下降的情况下,在 GFLOPs、FPS、参数数量以及模型权重上具有相对优势。 经过实验验证,模型检测精度保持 98. 3% 的同时能够更方便更容易部署在性能不高的移动设备。 该文为基于 YOLOv5n 的寄生卵检测和分类任务在实用性方面提供了一种有效的解决方案。
- Abstract:
-
Many current target detection models are too complex to implement parasitic egg detection and classification tasks on mobile devices. Therefore,we study an algorithm that integrates improved YOLOv5n and channel pruning. YOLOv5 is selected because of its lightweight and high accuracy,which can achieve the experimental purpose in this paper. The fusion C3 _Faster module and RepConv heavy parameterization module are used to replace all C3 modules in BackBone of YOLOv5n and some convolutional modules in Neck network. C3_Faster module reduces convolutional operations by PConv to speed up network model inference. The RepConv heavy pa-rameterization module implements multi-branch structure to enhance feature extraction capability in the training stage,and single-branch structure to accelerate detection speed in the verification stage. Meanwhile,sparse training and channel pruning are carried out on the improved YOLOv5n model. By reducing the redundant channels in the model,the complexity of the model is reduced,the number of pa-rameters is reduced, the detection efficiency is improved and the model weight is reduced. Compared with YOLOv5n, YOLOv5s, YOLOv7-tiny,YOLOv8n and SSD target detection algorithms, the proposed method has comparative advantages in GFLOPs, FPS, number of parameters and model weight under the condition of slightly reduced detection accuracy. After experimental verification,the model detection accuracy remains 98. 3% ,and it can be more convenient and easy to deploy on mobile devices with low performance.We provide an effective solution for the parasitic egg detection and classification task based on YOLOv5n in terms of practicality.
更新日期/Last Update:
2025-02-10