基于改进YOLOv9的群养猪身份识别模型

doi:10.13304/j.nykjdb.2025.0091

摘要/Abstract

摘要：

猪攻击过程中会产生身体形变、遮挡等因素，从而导致猪身份难以识别。提出一种基于改进YOLOv9的深度学习算法识别攻击状态下猪身份。从标记的600段1 s攻击视频中产生18 000帧图像作为数据集。首先，采用DualConv替换YOLOv9网络的下采样，在保持精度基础上降低计算量；然后，融合双向特征金字塔改进YOLOv9的颈部网络部分，以提升模型在攻击场景下的特征提取能力；接着，在主干网络的RepNCSPELAN4层后引入局部自注意力机制，以增强模型捕捉局部特征的能力；最后，采用改进的YOLOv9识别猪身份。结果表明，改进的YOLOv9模型识别猪身份的平均精度达93.6%，较基准模型提高3.7百分点，检测速度达31.58 帧·s^-1。以上表明，改进的YOLOv9算法能有效提升攻击场景下猪身份的识别精度，有助于将攻击识别从群体级细化为个体级。

关键词: 猪身份识别, 深度学习, YOLOv9, 双向特征金字塔网络, 自注意力机制

Abstract:

Body deformation， occlusion and other factors will generate during pig aggression， and this makes it difficult to recognize the identity of pigs. This study proposed a deep learning algorithm based on improved YOLOv9 to recognize pig identities under aggression situation. 18 000 frames were generated from 600 labelled 1 s aggressive videos as the dataset. Firstly， DualConv was used to replace the subsampling of YOLOv9 network， which reduced the computation on the base of maintaining the accuracy. Secondly， the neck part of YOLOv9 was improved by integrating bidirectional feature pyramid network （Bi-FPN） to improve the feature extraction capability of the model under aggression scene. Then， the partial self-attention （PSA） mechanism was introduced behind RepNCSPELAN4 layer of backbone to enhance the ability of model to capture local features. Finally， the improved YOLOv9 was used to recognize pig identities. The results showed that the average accuracy of this improved YOLOv9 model was 93.6%， which was 3.7 percentage point higher than that of the benchmark model， and the detecting speed was 31.58 f·s^-1. Above results indicated that the improved YOLOv9 could effectively improve the identification accuracy of pigs in aggression scene， and helped refine the aggression recognition from the group level to the individual level.

Key words: pig identification, deep learning, YOLOv9, bidirectional feature pyramid network（Bi-FPN）, self-attention mechanism

中图分类号:

S126

陈晨, 刘浩然, NORTON Tomas. 基于改进YOLOv9的群养猪身份识别模型[J]. 中国农业科技导报, 2025, 27(10): 134-143.

Chen CHEN, Haoran LIU, Tomas NORTON. Model for Identification of Group-housed Pigs Based on Improved YOLOv9[J]. Journal of Agricultural Science and Technology, 2025, 27(10): 134-143.

图/表 8

参考文献 19

[1]	郭婧驰, 张明源. 经济政策稳定性对我国生猪产业链价格的影响[J].经济纵横, 2021(1): 98-110.
	GUO J C, ZHANG M Y. The impact of economic policy stability on the price of China’s pig industry chain [J]. Econ. Rev. J., 2021(1): 98-110.
[2]	MA C, DENG M H, YIN Y L. Pig face recognition based on improved YOLOv4 lightweight neural network [J]. Inf. Process. Agric., 2024, 11(3): 356-371.
[3]	CHEN C, ZHU W X, STEIBEL J, et al.. Recognition of aggressive episodes of pigs based on convolutional neural network and long short-term memory [J/OL].Comput. Electron. Agric., 2020,169:105166 [2025-01-06]. .
[4]	CHEN C, ZHU W X, MA C H, et al.. Image motion feature extraction for recognition of aggressive behaviors among group-housed pigs [J]. Comput. Electron. Agric., 2017, 142: 380-387.
[5]	CHEN C, ZHU W X, GUO Y Z, et al.. A kinetic energy model based on machine vision for recognition of aggressive behaviours among group-housed pigs [J]. Livest. Sci., 2018,218: 70-78.
[6]	CHEN C, ZHU W X, NORTON T. Behaviour recognition of pigs and cattle:journey from computer vision to deep learning [J/OL]. Comput. Electron. Agric., 2021,187:106255 [2025-01-06]. .
[7]	TZANIDAKIS C, SIMITZIS P, ARVANITIS K, et al.. An overview of the current trends in precision pig farming technologies [J/OL]. Livest. Sci., 2021,249:104530 [2025-01-06]. .
[8]	VOULODIMOS A S, PATRIKAKIS C Z, SIDERIDIS A B, et al.. A complete farm management system based on animal identification using RFID technology [J]. Comput. Electron. Agric., 2010,70(2): 380-388.
[9]	HUANG W J, ZHU W X, MA C H, et al.. Identification of group-housed pigs based on Gabor and local binary pattern features [J]. Biosyst. Eng., 2018,166: 90-100.
[10]	BERGQVIST A S, FORSBERG F, ELIASSON C, et al.. Individual identification of pigs during rearing and at slaughter using microchips [J]. Livest. Sci., 2015,180: 233-236.
[11]	KASHIHA M, BAHR C, OTT S, et al.. Automatic identification of marked pigs in a pen using image pattern recognition [J].Comput. Electron. Agric., 2013, 93: 111-120.
[12]	GUO Y Z, ZHU W X, JIAO P P, et al.. Foreground detection of group-housed pigs based on the combination of mixture of Gaussians using prediction mechanism and threshold segmentation [J]. Biosyst. Eng., 2014,125: 98-104.
[13]	HANSEN M F, SMITH M L, SMITH L N,et al.. Towards on-farm pig face recognition using convolutional neural networks [J]. Comput. Ind., 2018, 98: 145-152.
[14]	WANG Z Y, LIU T H. Two-stage method based on triplet margin loss for pig face recognition [J/OL]. Comput. Electron. Agric., 2022,194:106737 [2025-01-06]. .
[15]	李勇,王杨,方夏,等.基于改进BiFPN的微特电机电枢表面缺陷检测方法[J].机床与液压, 2022, 50(6): 1-8.
	LI Y, WANG Y, FANG X, et al.. Micro-motor armature surface defect detection method based on improved BiFPN [J]. Mach. Tool Hydraul., 2022, 50(6): 1-8.
[16]	WANG C Y, YEH I H, MARK LIAO H Y. YOLOv9:learning what you want to learn using programmable gradient information [C]//Computer Vision-ECCV 2024.Cham:Springer Nature Switzerland, 2024: 1-21.
[17]	ZHONG J, CHEN J, MIAN A. DualConv:dual convolutional kernels for lightweight deep neural networks [J]. IEEE Trans. Neural Netw Learn. Syst., 2023, 34(11): 9528-9535.
[18]	WANG A, CHEN H, LIU L H, et al.. Yolov10: real-time end-to-end object detection [J]. Adv. Neural Inform. Process. Syst., 2024, 37: 107984-108011.
[19]	马娜, 徐苗.基于卷积神经网络的猪只个体身份识别研究[J].计算机时代, 2022(4): 51-54.
	MA N, XU M. Research on pig individual identity based on convolutional neural networks [J]. Comput. Era, 2022(4): 51-54.

融合方法 Fusion method			平均精度均值 mAP/%	平均精度AP/%								检测速度/（帧·s^-1） Detect speed/（f·s^-1）
DualConv	Bi-FPN	PSA	平均精度均值 mAP/%	Pig1	Pig2	Pig3	Pig4	Pig5	Pig6	Pig7	Pig8	检测速度/（帧·s^-1） Detect speed/（f·s^-1）
×	×	×	89.9	94.7	95.3	84.4	97.5	92.4	91.2	70.2	93.3	34.34
√	×	×	90.2	92.5	97.6	79.7	98.8	91.6	92.3	72.5	96.9	35.03
√	√	×	91.4	95.5	98.3	81.9	98.7	90.8	91.5	78.3	96.2	31.77
√	×	√	91.7	92.1	97.3	82.7	98.4	90.5	91.8	84.1	96.5	34.14
√	√	√	93.6	94.8	97.7	86.5	98.1	93.2	94.1	87.8	96.9	31.58

融合方法 Fusion method			平均精度均值 mAP/%	平均精度AP/%								检测速度/（帧·s^-1） Detect speed/（f·s^-1）
DualConv	Bi-FPN	PSA	平均精度均值 mAP/%	Pig1	Pig2	Pig3	Pig4	Pig5	Pig6	Pig7	Pig8	检测速度/（帧·s^-1） Detect speed/（f·s^-1）
×	×	×	89.9	94.7	95.3	84.4	97.5	92.4	91.2	70.2	93.3	34.34
√	×	×	90.2	92.5	97.6	79.7	98.8	91.6	92.3	72.5	96.9	35.03
√	√	×	91.4	95.5	98.3	81.9	98.7	90.8	91.5	78.3	96.2	31.77
√	×	√	91.7	92.1	97.3	82.7	98.4	90.5	91.8	84.1	96.5	34.14
√	√	√	93.6	94.8	97.7	86.5	98.1	93.2	94.1	87.8	96.9	31.58

模型Model	精确度P/%	召回率R/%	平均精度均值mAP/%	检测速度/（帧·s^-1） Detect speed/（f·s^-1）
Faster R-CNN	69.8	68.4	70.3	23.76
SSD	80.7	75.8	81.6	24.94
YOLOv5	77.2	70.9	78.9	33.54
YOLOv7	84.1	79.6	85.2	32.52
YOLOv8	88.8	80.7	89.4	30.14
YOLOv10	88.1	82.0	89.7	31.13
改进模型Improveed model	93.4	88.3	93.6	31.58

模型Model	精确度P/%	召回率R/%	平均精度均值mAP/%	检测速度/（帧·s^-1） Detect speed/（f·s^-1）
Faster R-CNN	69.8	68.4	70.3	23.76
SSD	80.7	75.8	81.6	24.94
YOLOv5	77.2	70.9	78.9	33.54
YOLOv7	84.1	79.6	85.2	32.52
YOLOv8	88.8	80.7	89.4	30.14
YOLOv10	88.1	82.0	89.7	31.13
改进模型Improveed model	93.4	88.3	93.6	31.58