四川轻化工大学学报（自然科学版）2026，Vol.39Issue(2)：84-95,12.DOI:10.11863/j.suse.2026.02.08
基于MBD-RTDETR的多场景生猪计数技术
Multi-scenario Pig Counting Technology Based on MBD-RTDETR
摘要
Abstract
To facilitate the digital management of swine assets,an improved algorithm based on the MBD-RTDETR model is proposed for counting pigs across various scenarios.Built upon the RTDETR framework,this algorithm incorporates the MobileNetV4 backbone network for optimization,thereby reducing model complexity and parameter count while enhancing operational efficiency and deployment flexibility.Furthermore,the introduction of a BiFPN structure strengthens the model's capability to capture multi-scale features,while the integration of a DAttention mechanism improves overall accuracy.To further improve performance,the algorithm also combines masking and ByteTrack tracking algorithms to ultimately achieve the identification of the number of pigs inside the pen and through the passageway.Compared to the baseline RTDETR model,MBD-RTDETR demonstrates improvements across multiple performance metrics:parameter count is reduced by 48.72%,computational load by 36.03%,and model size by 48.31%;concurrently,the P(Precision)increases by 1.45 percentage points,the F1 score improves by 0.42,and the AP(Average Precision)rises by 1.07 percentage points.The improved MBD-RTDETR model serves as a valuable technical reference for smart swine farming applications.关键词
生猪计数/RTDETR/MobileNetV4/BiFPN/DAttention/智慧养殖Key words
pig counting/RTDETR/MobileNetV4/BiFPN/DAttention/smart farming分类
信息技术与安全科学引用本文复制引用
钟长华,宋弘,吴浩,江俊卓,俞婧婧..基于MBD-RTDETR的多场景生猪计数技术[J].四川轻化工大学学报（自然科学版）,2026,39(2):84-95,12.基金项目
四川省科技厅项目(2022YFS08 （）
2022ZHCG0035) （）
智能感知与控制四川省重点实验室(2023RYY06) （2023RYY06）
四川轻化工大学研究生创新基金项目(Y2023296) （Y2023296）
