Work

Lameness detection in dairy cows from overhead view: high-precision keypoint localization and multi-feature fusion classification

Year: 2025

Type: article

Abstract: Introduction: Detecting lameness in dairy cows from an overhead view can effectively avoid occlusion caused by farm facilities or other animals, while suspended detection devices enable parallel monitoring without disturbing natural behaviors. However, existing methods from this perspective still face challenges in accuracy and generalization, largely due to the subtlety of back movement features and individual variability. To address these limitations, this study explores an overhead-view lameness detection approach based on RGB-D data. Methods: We developed a high-precision keypoint detection method for the cow's back that models long-range spatial dependencies and optimizes structural representation. On this basis, six lameness-related features were designed to capture posture and motion abnormalities, including four newly proposed indices. Their correlation in classifying sound, mildly lame, and severely lame cows was systematically analyzed. To further enhance robustness, the Gini importance index from Random Forest combined with a permutation importance correction method (PIMP) was applied to construct an unbiased feature selection framework. Results: Experimental results demonstrate that the proposed keypoint detection network achieved a [email protected] of 100.00% and an average precision of 95.89%, significantly outperforming the baseline model. In feature-based classification, back curvature, movement asymmetry index, and vertical oscillations of the back and head exhibited strong discriminative ability. Using multi-feature fusion, the lameness detection model attained an overall accuracy of 91.00%. Discussion: These findings indicate that overhead RGB-D imaging, combined with precise keypoint detection and feature fusion, provides a reliable strategy for accurate lameness detection in dairy cows. The proposed method offers valuable theoretical and technical support for health monitoring and intelligent management in modern dairy farming.

Source: Frontiers in Veterinary Science

DOI: 10.3389/fvets.2025.1675181

Authors: Weijun Duan, Fang Wang, Honghui Li, Na Liu, Xueliang Fu

Institution: Inner Mongolia Agricultural University

Language: English

Topic: Smart Agriculture and AI

Subfield: Plant Science

Field: Agricultural and Biological Sciences

Domain: Life Sciences

SDG: Zero hunger

Open Access status: gold

Estimated APC paid: $3,801