农业机械学报2025，Vol.56Issue(10)：575-584,10.DOI:10.6041/j.issn.1000-1298.2025.10.052
基于无人机图像和MLL-YOLO v10s的草原放牧羊只实时检测模型
Real-time Detection Model for Grazing Sheep in Grassland Based on UAV Imagery and MLL-YOLO v10s
摘要
Abstract
Aiming to meet the demand of herdsmen in Inner Mongolia grassland pastures for accurate real-time monitoring and management of large flocks of free-range sheep,a high-precision and lightweight real-time unmanned aerial vehicle(UAV)remote sensing target detection model named MLL-YOLO v10s(MobileNetV4 LSKA LSCD-YOLO v10s)was proposed.This model enabled real-time detection of individual sheep in large flocks from the high-altitude perspective of UAVs.To address the challenges of difficult sheep detection and poor real-time performance caused by densely packed and mutually occluded sheep,the following improvements were made based on the YOLO(you only look once)v10 model.MobileNetV4 was employed as the backbone network to reduce the number of model parameters and enhance computational efficiency.The large separable kernel attention(LSKA)module was introduced to strengthen the model's ability to capture features of small targets.A lightweight shared convolutional detection head(LSCD)was designed to reduce computational redundancy through weight sharing and improve the computational efficiency of the model.Compared with the YOLO series,faster regions with convolutional neural networks(Faster R-CNN),and other classic network models,the improved MLL-YOLO v10s model achieved a mean average precision(mAP)of 93.6％on the test set,which was 3.4 percentage points higher than that of the baseline model.It had an average frame rate of 135 f/s and only 1.268 × 107 parameters.In densely occluded scenarios,the false-negative rate was significantly reduced.The model's size and computational requirements were superior to those of mainstream single/dual-stage target detection algorithms.The proposed MLL-YOLO v10s model demonstrated stronger robustness in detecting densely aggregated and partially occluded sheep in UAV aerial photography scenarios.It also had obvious advantages in terms of the number of parameters and computational requirements.This model provided support for the combined application of edge computing devices and UAVs,offering an effective real-time detection method for UAV-based sheep flock inspections in natural pastures.关键词
羊只检测/YOLO v10/无人机/小目标/MobileNetV4/轻量化Key words
sheep detection/YOLO v10/UAV/small object/MobileNetV4/lightweight分类
信息技术与安全科学引用本文复制引用
张东彦,叶佳炜,郭阳阳,胡根生,李威风,唐晶磊,韩冬..基于无人机图像和MLL-YOLO v10s的草原放牧羊只实时检测模型[J].农业机械学报,2025,56(10):575-584,10.基金项目
内蒙古自治区科技计划项目(2022YFSJ0039) （2022YFSJ0039）
