基于改进YOLOv4模型的群养生猪姿态检测

doi:10.3969/j.issn.1004-1524.2023.01.23

浙江农业学报 ›› 2023, Vol. 35 ›› Issue (1): 215-225.DOI: 10.3969/j.issn.1004-1524.2023.01.23

基于改进YOLOv4模型的群养生猪姿态检测

李斌¹(), 刘东阳¹, 时国龙¹, 慕京生², 徐浩然¹, 辜丽川¹, 焦俊¹^,^*()

1.安徽农业大学信息与计算机学院,安徽合肥 230036
2.蒙城县京徽蒙农业科技发展有限公司,安徽亳州 233524

收稿日期:2021-06-29 出版日期:2023-01-25 发布日期:2023-02-21
通讯作者: *焦俊,E-mail:jiaojun2000@sina.com.cn
作者简介:李斌(1996—),男,安徽阜阳人,硕士研究生,研究方向为模式识别。E-mail:735438610@qq.com
基金资助:
安徽省科技重大专项(201903a06020009)

Pig posture detection based on improved YOLOv4 model

LI Bin¹(), LIU Dongyang¹, SHI Guolong¹, MU Jingsheng², XU Haoran¹, GU Lichuan¹, JIAO Jun¹^,^*()

1. School of Information and Computer, Anhui Agricultural University, Hefei 230036, China
2. Mengcheng Jinghuimeng Agricultural Science and Technology Development Co., Ltd., Bozhou 233524, Anhui, China

Received:2021-06-29 Online:2023-01-25 Published:2023-02-21

摘要/Abstract

摘要：

为了提升猪舍环境下生猪姿态检测的速度和性能,在YOLOv4模型的基础上提出一种改进的Mini_YOLOv4模型。首先,该模型将YOLOv4的特征提取网络改为轻量级的MobileNetV3网络结构,以降低模型参数量;其次,在检测网络的CBL_block1、CBL_block2模块中使用深度可分离卷积代替传统卷积,避免了复杂模型导致的内存不足和高延迟问题;最后,将原YOLOv4网络每个尺度的最后一层3×3卷积改为Inception网络结构,以提高模型在生猪姿态检测上的准确率。应用上述模型,对生猪的站立、坐立、腹卧、趴卧和侧卧5类姿态进行识别。结果显示, Mini_YOLOv4模型较YOLOv4模型在检测精度上提升了4.01百分点,在检测速度上提升近1倍,在保证识别精度的同时提升了实时性,可为生猪行为识别提供技术参考。

关键词: YOLOv4模型, MobileNetV3网络, 生猪姿态检测

Abstract:

In order to improve the speed and performance of pig posture detection in piggery environment, an improved Mini_YOLOv4 model was proposed in the present assay based on YOLOv4 model. Firstly, the feature extraction network of YOLOv4 was changed into lightweight MobileNetV3 network structure to reduce the amount of model parameters. Secondly, the deep separable convolution was used to replace the traditional convolution in the CBL_block1 and CBL_block2 modules of the detection network to avoid the memory shortage and high delay problems caused by the complex model. Finally, the 3×3 convolution at each scale of the original YOLOv4 network was changed into the Inception network structure to improve the accuracy of the model. The performance of the above models was evaluated in the detection of five types of postures of pigs, i.e. stand, sit, lie, ventral and lateral. It was shown that, compared with the YOLOv4 model, the detection accuracy of the proposed Mini_YOLOv4 model was improved by 4.01 percentage points, and its detection speed was almost doubled. In summary, the proposed model could improve the real-time performance and ensure the accuracy of pig posture detection, which could provide technical references for pig behavior identification.

Key words: YOLOv4 model, MobileNetV3 network, pig posture detection

中图分类号:

TP391.4

李斌, 刘东阳, 时国龙, 慕京生, 徐浩然, 辜丽川, 焦俊. 基于改进YOLOv4模型的群养生猪姿态检测[J]. 浙江农业学报, 2023, 35(1): 215-225.

LI Bin, LIU Dongyang, SHI Guolong, MU Jingsheng, XU Haoran, GU Lichuan, JIAO Jun. Pig posture detection based on improved YOLOv4 model[J]. Acta Agriculturae Zhejiangensis, 2023, 35(1): 215-225.

图/表 14

图1 YOLOv4网络结构示意图

Fig.1 Schematic diagram of YOLOv4 network structure

表1 MobileNetV3的网络结构

Table 1 Structure of MobileNetV3 network

类型 Type	滤波器 Filters	Exp size	SE	激活函数 Activation function	步长 Stride	输出 Output
Conv2D	16	—	否No	HS	2	208×208
Bneck,3×3	16	16	否No	HS	1	208×208
Bneck,3×3	24	64	否No	RE	2	104×104
Bneck,3×3	24	72	否No	HS	1	104×104
Bneck,5×5	40	72	是Yes	RE	2	52×52
Bneck,5×5	40	120	是Yes	RE	1	52×52
Bneck,5×5	40	120	是Yes	RE	1	52×52
Bneck,3×3	80	240	否No	HS	2	26×26
Bneck,3×3	80	200	否No	HS	1	26×26
Bneck,3×3	80	184	否No	HS	1	26×26
Bneck,3×3	80	184	是Yes	HS	1	26×26
Bneck,3×3	112	480	是Yes	HS	1	26×26
Bneck,3×3	112	672	是Yes	HS	1	26×26
Bneck,5×5	160	672	是Yes	HS	2	13×13
Bneck,5×5	160	960	是Yes	HS	1	13×13
Bneck,5×5	160	960	否No	HS	1	13×13

图2 DBH_block1 (a)和DBH_block2 (b)的网络结构示意图

Fig.2 Schematic diagram of DBH_block1 (a) and DBH_block2 (b)

图3 原版Inception结构(a)与改进后的Inception结构(b)示意图

Fig.3 Schematic diagram of original (a) and improved inception (b) structure

图4 整体改进后的Mini_YOLOv4网络结构示意图

Fig.4 Schematic diagram of improved Mini_YOLOv4 network structure

图5 生猪5类姿态的对应图像

Fig.5 Images of five types of daily postures of pigs

图6 YOLOv4(a)和Mini_YOLOv4(b)模型的损失值曲线 train-loss,训练损失值;val-loss,验证损失值。

Fig.6 Loss curves of YOLOv4 (a) and mini_YOLOv4 (b) train-loss, Loss in training; val-loss, Loss in validation.

图7 YOLOv4(a)和Mini_YOLOv4(b)模型的识别结果对比

Fig.7 Comparison of identification results of YOLOv4 (a) and Mini_YOLOv4 (b) models

表2 不同IoU阈值下各模型的均值平均精度(mAP)

Table 2 Mean average precision (mAP) of different models under different IoU thresholds %

模型 Model	不同IoU阈值下的mAP mAP under different IoU thresholds
模型 Model	0.5	0.6	0.7	0.75
YOLOv4	69.66	66.71	58.34	52.94
Mini_YOLOv4	73.67	73.67	73.00	71.35

表3 不同模型的检测精度与检测速度

Table 3 Detection accuracy and speed of different models

模型 Model	不同姿态下的检测精度Precision under different postures/%					检测速度 Speed/(frame·s^-1)
模型 Model	站立Stand	趴卧Lie	坐立Sit	腹卧Ventral	侧卧Lateral	检测速度 Speed/(frame·s^-1)
YOLOv4	66.17	69.25	71.15	64.82	76.89	33.98
Mini_YOLOv4	68.59	70.87	74.10	71.27	83.51	65.62

表4 不同模型的召回率与F1值

Table 4 Recall and F1 values of different models

模型 Model	不同姿态下的召回率Recall under different postures/%					不同姿态下的F1值F1 under different postures/%
模型 Model	站立 Stand	趴卧 Lie	坐立 Sit	腹卧 Ventral	侧卧 Lateral	站立 Stand	趴卧 Lie	坐立 Sit	腹卧 Ventral	侧卧 Lateral
YOLOv4	66.67	70.26	73.78	71.22	84.54	0.78	0.78	0.79	0.75	0.78
Mini_YOLOv4	69.57	71.28	74.22	72.20	83.51	0.80	0.81	0.85	0.82	0.91

图8 YOLOv4(a)和Mini_YOLOv4(b)模型定位结果对比

Fig.8 Comparison of positioning results of YOLOv4 (a) and Mini_YOLOv4 (b) models

图9 YOLOv4(a)、Mini_YOLOv4(b)模型的检测可靠性

Fig.9 Detection reliability of YOLOv4(a) and Mini_YOLOv4(b) models

图10 YOLOv4(a)和Mini_YOLOv4(b)模型的检测结果对比

Fig.10 Comparison of detection results of YOLOv4 (a) and Mini_YOLOv4 (b) models

参考文献 28

[1]	党亚男, 王芳, 田建艳, 等. 面向猪的姿态识别的特征优选方法研究[J]. 江苏农业科学, 2016, 44(3): 448-451.
	DANG Y N, WANG F, TIAN J Y, et al. Research on feature optimization method for pig pose recognition[J]. Jiangsu Agricultural Sciences, 2016, 44(3): 448-451. (in Chinese)
[2]	杨秋妹, 肖德琴, 张根兴. 猪只饮水行为机器视觉自动识别[J]. 农业机械学报, 2018, 49(6): 232-238.
	YANG Q M, XIAO D Q, ZHANG G X. Automatic pig drinking behavior recognition with machine vision[J]. Transactions of the Chinese Society for Agricultural Machinery, 2018, 49(6): 232-238. (in Chinese with English abstract)
[3]	嵇杨培, 杨颖, 刘刚. 基于可见光光谱和YOLOv2的生猪饮食行为识别[J]. 光谱学与光谱分析, 2020, 40(5): 1588-1594.
	JI Y P, YANG Y, LIU G. Recognition of pig eating and drinking behavior based on visible spectrum and YOLOv2[J]. Spectroscopy and Spectral Analysis, 2020, 40(5): 1588-1594. (in Chinese with English abstract)
[4]	TRAULSEN I, SCHEEL C, AUER W, et al. Using acceleration data to automatically detect the onset of farrowing in sows[J]. Sensors (Basel, Switzerland), 2018, 18(1): 170. DOI URL
[5]	王凯, 刘春红, 段青玲. 基于MFO-LSTM的母猪发情行为识别[J]. 农业工程学报, 2020, 36(14): 211-219.
	WANG K, LIU C H, DUAN Q L. Identification of sow oestrus behavior based on MFO-LSTM[J]. Transactions of the Chinese Society of Agricultural Engineering, 2020, 36(14): 211-219. (in Chinese with English abstract)
[6]	何屿彤, 李斌, 张锋, 等. 基于改进YOLOv3的猪脸识别[J]. 中国农业大学学报, 2021, 26(3): 53-62.
	HE Y T, LI B, ZHANG F, et al. Pig face recognition based on improved YOLOv3[J]. Journal of China Agricultural University, 2021, 26(3): 53-62. (in Chinese with English abstract)
[7]	焦俊, 王文周, 侯金波, 等. 基于改进残差网络的黑毛猪肉新鲜度识别方法[J]. 农业机械学报, 2019, 50(8): 364-371.
	JIAO J, WANG W Z, HOU J B, et al. Freshness identification of iberico pork based on improved residual network[J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50(8): 364-371. (in Chinese with English abstract)
[8]	邱洪涛, 孙裴, 侯金波, 等. 基于Caffe的猪肉新鲜度分级的设计与实现[J]. 江苏农业学报, 2019, 35(2): 461-468.
	QIU H T, SUN P, HOU J B, et al. Design and implementation of pork freshness grading based on Caffe[J]. Jiangsu Journal of Agricultural Sciences, 2019, 35(2): 461-468. (in Chinese with English abstract)
[9]	MARSOT M, MEI J Q, SHAN X C, et al. An adaptive pig face recognition approach using convolutional neural networks[J]. Computers and Electronics in Agriculture, 2020, 173: 105386. DOI URL
[10]	甘海明, 薛月菊, 李诗梅, 等. 基于时空信息融合的母猪哺乳行为识别[J]. 农业机械学报, 2020, 51(S1): 357-363.
	GAN H M, XUE Y J, LI S M, et al. Automatic sow nursing behaviour recognition based on spatio-temporal information fusion[J]. Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(S1): 357-363. (in Chinese with English abstract)
[11]	燕红文, 刘振宇, 崔清亮, 等. 基于改进Tiny-YOLO模型的群养生猪脸部姿态检测[J]. 农业工程学报, 2019, 35(18): 169-179.
	YAN H W, LIU Z Y, CUI Q L, et al. Detection of facial gestures of group pigs based on improved Tiny-YOLO[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(18): 169-179. (in Chinese with English abstract)
[12]	PSOTA E T, MITTEK M, PÉREZ L C, et al. Multi-pig part detection and association with a fully-convolutional network[J]. Sensors (Basel, Switzerland), 2019, 19(4): 852. DOI URL
[13]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition.Columbus, OH, USA: IEEE, 2014: 580-587.
[14]	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI PMID
[15]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[J]. Lecture Notes in Computer Science, 2016, 9905: 21-37.
[16]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: IEEE, 2016: 779-788.
[17]	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA: IEEE, 2017: 6517-6525.
[18]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. (2018-04-08) [2021-06-29]. https://arxiv.org/abs/1804.02767
[19]	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. (2020-04-23) [2021-06-29]. https://arxiv.org/abs/2004.10934
[20]	HOWARD A, SANDLER M, CHEN B, et al. Searching for MobileNetV3[C]// 2019 IEEE/CVF International Conference on Computer Vision (ICCV). October 27-November 2, 2019. Seoul:IEEE, 2019: 1314-1324.
[21]	CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA: IEEE, 2017: 1800-1807.
[22]	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: IEEE, 2016: 2818-2826.
[23]	WANG C Y, MARK LIAO H Y, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, WA, USA: IEEE, 2020: 1571-1580.
[24]	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA: IEEE, 2018: 8759-8768.
[25]	HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. (2017-04-17) [2021-06-29]. https://arxiv.org/abs/1704.04861
[26]	SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA. IEEE, 2018: 4510-4520.
[27]	TAN M X, CHEN B, PANG R M, et al. MnasNet: platform-aware neural architecture search for mobile[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach, CA, USA: IEEE, 2019: 2815-2823.
[28]	薛月菊, 朱勋沐, 郑婵, 等. 基于改进Faster R-CNN识别深度视频图像哺乳母猪姿态[J]. 农业工程学报, 2018, 34(9): 189-196.
	XUE Y J, ZHU X M, ZHENG C, et al. Lactating sow postures recognition from depth image of videos based on improved Faster R-CNN[J]. Transactions of the Chinese Society of Agricultural Engineering, 2018, 34(9): 189-196. (in Chinese with English abstract)

基于改进YOLOv4模型的群养生猪姿态检测

Pig posture detection based on improved YOLOv4 model

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 28

相关文章 12

编辑推荐

Metrics

本文评价

[1]	闫宁, 张晗, 董宏图, 康凯, 罗斌. 基于透射光和反射光图像同位分割的小麦品种识别方法研究[J]. 浙江农业学报, 2022, 34(3): 590-598.
[2]	林孝腾, 孙茜, 郑书河. 基于OpenCV的低能耗植物工厂照明系统设计与试验[J]. 浙江农业学报, 2022, 34(3): 626-635.
[3]	肖志云, 王伊凝. 基于RF-VR的紫丁香叶片叶绿素含量高光谱反演[J]. 浙江农业学报, 2021, 33(11): 2164-2173.
[4]	钟伟镇, 刘鑫磊, 杨坤龙, 李丰果. 基于Mask-RCNN的复杂背景下多目标叶片的分割和识别[J]. 浙江农业学报, 2020, 32(11): 2059-2066.
[5]	闫建伟, 苏小东, 赵源, 刘进平. 基于图像特征的小麦胚芽鞘识别[J]. 浙江农业学报, 2019, 31(2): 326-332.
[6]	张满囤, 单新媛, 于洋, 米娜, 阎刚, 郭迎春. 基于小波变换和改进KPCA的奶牛个体识别研究[J]. 浙江农业学报, 2017, 29(12): 2000-2008.
[7]	毕立恒. 基于叶片图像算法的植物种类识别方法研究[J]. 浙江农业学报, 2017, 29(12): 2142-2148.
[8]	张善文, 黄文准, 尤著宏. 基于物联网和深度卷积神经网络的冬枣病害识别方法[J]. 浙江农业学报, 2017, 29(11): 1868-1874.
[9]	李颀, 胡艺聪, 武付闯. 基于图像处理的育苗箱环境控制系统[J]. 浙江农业学报, 2017, 29(11): 1912-1919.
[10]	夏永泉, 王兵, 支俊, 黄海鹏, 孙静茹. 基于EM和K-means混合聚类方法的植物叶片病害区域自动提取[J]. 浙江农业学报, 2017, 29(8): 1384-1391.
[11]	吴庆岗, 张卫国, 常化文, 金保华, 刘朝霞. 基于梯度Hough变换的遮挡苹果目标定位[J]. 浙江农业学报, 2017, 29(6): 1009-1016.
[12]	张善文, 孔韦韦, 王震. 基于稀疏表示字典学习的植物分类方法[J]. 浙江农业学报, 2017, 29(2): 338-344.