基于SAM优化的饲喂目标实时识别方法

doi:10.12141/j.issn.1000-565X.240591

摘要/Abstract

摘要：

饲喂辅助机器人是推动畜牧业现代化转型的关键设备，饲喂目标的快速、准确识别是机器人实现智能推料的重要保证。匹配分割精度和运行效率是保证算法综合性能的关键步骤，也是识别算法的重要课题。针对现有奶牛饲喂目标识别方法存在分割精度和运行效率不匹配的问题，该文提出了一种基于分割大模型（SAM）优化的饲喂目标实时识别方法RTFTR。该方法首先在SAM-det架构基础上，通过轻量化图像编码器和目标检测器的参数，引入缓冲区队列的并行化设计方法来平衡各模块的运行效率，以提升推理速率；然后利用HQ形符增强特征空间的解码能力，优化设计掩码解码器，并采用针对饲喂目标的分阶段训练，以提高分割精度。实验结果表明：所提方法在提高分割精度的前提下保证了推理速率；在奶牛饲喂目标识别中，奶牛分割精度达98.7%，饲料分割精度达96.4%，料槽分割精度达99.2%，整体平均分割精度达98.1%，运行速率为52.9 f/s，满足养殖场复杂环境和机器人计算资源限制下对奶牛饲喂目标识别方法的高精度、高效率的应用需求。

关键词: 饲喂辅助机器人, 分割大模型, 奶牛饲喂, 目标识别, 分割精度

Abstract:

Feeding-assistance robots are key equipment in promoting the modernization and transformation of animal husbandry. The rapid and accurate identification of feeding targets is essential for enabling intelligent feed-pushing, while balancing segmentation accuracy and operational efficiency is crucial for ensuring the overall performance of recognition algorithms—an important topic in the field of intelligent livestock management. To address the mismatch between segmentation accuracy and processing efficiency in current dairy cow feeding target recognition methods, this paper proposed a real-time feeding target recognition method (RTFTR) based on an optimized Segment Anything Model (SAM). Built on the SAM-det architecture, RTFTR first introduces lightweight image encoder and object detector, along with a parallelized buffer queue design, to balance the operational efficiency of each module and enhance inference speed. It then employs a High-Quality (HQ) token mechanism to enhance the feature space decoding capacity, optimizes the mask decoder, and applies stage-wise training tailored to feeding targets to improve segmentation accuracy. Experimental results show that the proposed method ensures inference efficiency while enhancing segmentation accuracy. In the task of cow feeding target recognition, the method achieves a segmentation accuracy of 98.7% for cows, 96.4% for feed, 99.2% for bunk, with an overall average accuracy of 98.1%, and a processing speed of 52.9 f/s, meeting the application requirements for cow feeding target recognition in complex environments and limited robotic computational resources.

Key words: feeding-assistance robot, segment anything model, cow feeding, target recognition, segmentation accuracy

中图分类号:

TP 391

张勤, 翁凯航. 基于SAM优化的饲喂目标实时识别方法[J]. 华南理工大学学报(自然科学版), 2025, 53(7): 60-69.

ZHANG Qin, WENG Kaihang. Real-Time Feeding Target Recognition Method Based on SAM Optimization[J]. Journal of South China University of Technology(Natural Science Edition), 2025, 53(7): 60-69.

图/表 7

图1

表1

图2

图3

表2

饲喂目标分割实验结果对比"

方法	平均分割精度/%	奶牛分割精度/%	饲料分割精度/%	料槽分割精度/%	推理速率/（f‧s^-1）	$Q S P E$
Mask R-CNN	95.9	98.6	90.7	98.4	7.4	26.64
YOLACT	94.4	98.5	86.5	98.2	20.9	44.42
SOLO	93.3	98.9	82.9	98.2	16.2	38.88
SOLOv2	92.6	97.7	94.0	86.2	17.9	40.71
YOLOv8-seg-m	90.0	96.2	85.8	89.7	40.6	60.45
SAM（H）	98.5	98.7	97.1	99.2	0.9	9.42
SAM（B）	98.3	98.7	96.8	99.1	2.1	14.37
HQSAM	98.5	98.7	97.0	99.2	2.1	14.38
EfficientSAM	98.1	98.7	96.3	99.2	10.4	31.94
RTFTR	98.1	98.7	96.4	99.2	19.5	43.74
RTFTR-TensorRT	98.0	98.7	95.8	99.2	52.9	72.00

表2

图4

表3

参考文献 30

[1]	张勤，任海林，胡嘉辉．基于信息融合的智能推料机器人设计与试验［J］．农业机械学报，2022，54（6）：78-84，93．
	ZHANG Qin， REN Hailin， HU Jiahui ．Design and experiment of intelligent feed-pushing robot based on information fusion［J］．Transactions of the Chinese Society for Agricultural Machinery，2022，54（6）：78-84，93．
[2]	张勤，胡嘉辉，任海林．饲喂辅助机器人的智能推料方法与试验研究［J］．华南理工大学学报（自然科学版），2022，50（6）：111-120．
	ZHANG Qin， HU Jiahui， REN Hailin ．Intelligent pushing method and experiment of feeding assistant robot［J］．Journal of South China University of Technology（Natural Science Edition），2022，50（6）：111-120．
[3]	ALBRIGHT J L ．Feeding behavior of dairy cattle［J］．Journal of Dairy Science，1993，76（2）：485-498.
[4]	STANKOVSKI S， OSTOJIC G， SENK I，et al ．Dairy cow monitoring by RFID［J］．Scientia Agricola，2012，69：75-80．
[5]	PORTO S M C， ARCIDIACONO C， GIUMMARRA A，et al ．Localisation and identification performances of a real-time location system based on ultra wide band technology for monitoring and tracking dairy cow beha-viour in a semi-open free-stall barn［J］．Computers and Electronics in Agriculture，2014，108：221-229．
[6]	姜威，郭庆，司陈，等．一种用于奶牛定位的装置和方法：CN201310228400.3［P］．2013-09-18．
[7]	DHANYA V G， SUBEESH A， KUSHWAHA N L，et al ．Deep learning based computer vision approaches for smart agricultural applications［J］．Artificial Intelligence in Agriculture，2022，6：211-229．
[8]	DIWAN T， ANIRUDH G， TEMBHURNE J V ．Object detection using YOLO：challenges，architectural succe-ssors，datasets and applications［J］．Multimedia Tools and Applications，2023，82（6）：9243-9275．
[9]	KUAN C Y， TSAI Y C， HSU J T，et al ．An imaging system based on deep learning for monitoring the feeding behavior of dairy cows［C］∥ Proceedings of 2019 ASABE Annual International Meeting．Boston：American Society of Agricultural and Biological Engineers，2019：1/1-9．
[10]	JIANG B， WU Q， YIN X，et al ．FLYOLOv3 deep learning for key parts of dairy cow body detection［J］．Computers and Electronics in Agriculture，2019，166：104982/1-8．
[11]	BEZEN R， EDAN Y， HALACHMI I ．Computer vision system for measuring individual cow feed intake using RGB-D camera and deep learning algorithms［J］．Computers and Electronics in Agriculture，2020，172：105345/1-11．
[12]	WANG J， ZHANG X， GAO G，et al ．Open pose mask R-CNN network for individual cattle recognition［J］．IEEE Access，2023，11：113752-113768．
[13]	LI X， GAO R， LI Q，et al ．Multi-target feeding-behavior recognition method for cows based on improved RefineMask［J］．Sensors，2024，24（10）：2975/1-19．
[14]	BOLYA D， ZHOU C， XIAO F，et al ．YOLACT：real-time instance segmentation［C］∥ Proceedings of 2019 IEEE/CVF International Conference on Computer Vision．Seoul：IEEE，2019：9156-9165．
[15]	YANG G， LI R， ZHANG S，et al ．Extracting cow point clouds from multi-view RGB images with an improved YOLACT++ instance segmentation［J］．Expert Systems with Applications，2023，230：120730/1-14．
[16]	KIRILLOV A， MINTUN E， RAVI N，et al ．Segment anything［C］∥ Proceedings of 2023 IEEE/CVF International Conference on Computer Vision．Paris：IEEE，2023：3992-4003．
[17]	ZHANG J ．Research on derived tasks and realistic applications of segment anything model：a literature review［J］．Frontiers in Computing and Intelligent Systems，2023，5（2）：116-119．
[18]	YANG X， DAI H， WU Z，et al ．An innovative segment anything model for precision poultry monitoring ［J］．Computers and Electronics in Agriculture，2024，222：109045/1-12．
[19]	LI Q， ZHAO J， BAI D，et al ．Cattle body size measurement system based on dual-position cameras［C］∥ Proceedings of 2024 International Conference on Computer Vision and Deep Learning．Changsha：ACM，2024：47/1-6．
[20]	CHEN K， LIU C， CHEN H，et al ．RSPrompter：learning to prompt for remote sensing instance segmentation based on visual foundation model［J］．IEEE Tran-sactions on Geoscience and Remote Sensing，2024，62：4701117/1-17．
[21]	DOSOVITSKIY A， BEYER L， KOLESNIKOV A，et al ．An image is worth 16　×　16 words：transformers for image recognition at scale［C］∥ Proceedings of the 9th International Conference on Learning Representations．［S.l.］：OpenReview.net，2021：1-21．
[22]	XIONG Y， VARADARAJAN B， WU L，et al ．EfficientSAM：leveraged masked image pretraining for efficient segment anything［C］∥ Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition．Seattle：IEEE，2024：16111-16121．
[23]	叶峰，陈彪，赖乙宗．基于特征空间嵌入的对比知识蒸馏算法［J］．华南理工大学学报（自然科学版），2023，51（5）：13-23．
	YE Feng， CHEN Biao， LAI Yizong ．Contrastive knowledge distillation method based on feature space embedding［J］．Journal of South China University of Technology （Natural Science Edition），2023，51（5）：13-23．
[24]	KE L， YE M， DANELLJAN M，et al ．Segment anything in high quality［C］∥ Advances in Neural Information Processing Systems 36：37th Conference on Neural Information Processing Systems．San Diego：Neural Information Processing Systems Foundation，Inc.，2023：29914-29934．
[25]	LYU C， ZHANG W， HUANG H，et al ．RTMDet：an empirical study of designing real-time object detectors ［EB/OL］．（2022-12-16）［2024-11-25］．．
[26]	LI Y， MAO H， GIRSHICK R，et al ．Exploring plain vision transformer backbones for object detection［C］∥Proceedings of the 17th European Conference on Computer Vision．Tel Aviv：Springer，2022：280-296．
[27]	TIAN Y， SU D， LAURIA S，et al ．Recent advances on loss functions in deep learning for computer vision ［J］．Neurocomputing，2022，497：129-158．
[28]	HE K， GKIOXARI G， DOLLÁR P，et al ．Mask R-CNN［C］∥ Proceedings of 2017 IEEE International Conference on Computer Vision．Venice：IEEE，2017：2961-2969．
[29]	WANG X， KONG T， SHEN C，et al ．SOLO： segmenting objects by locations［C］∥ Proceedings of the 16th European Conference on Computer Vision．Glasgow：Springer，2020：649-665.
[30]	WANG X， ZHANG R， KONG T，et al ．SOLOv2：dynamic and fast instance segmentation［C］∥ Advances in Neural Information Processing Systems 33：34th Conference on Neural Information Processing Systems．San Diego：Neural Information Processing Systems Foundation，Inc.， 2020：17721-17732．