Answers are grounded in this paper and cite their sources.
Comprehensive review analyzes multimodal fusion strategies in livestock farming, suggesting solutions for ongoing challenges.
• Fills a critical gap with the first comprehensive survey of PLF multi-modal fusion. • Primary sensing modalities and their standalone limitations are analyzed. • A novel taxonomy categorizes fusion by data homogeneity and integration strategies. • Delivers a curated benchmark of open-access multi-modal PLF datasets. • Pioneers actionable solutions for navigating key challenges in multi-modal PLF. Technology-driven precision livestock farming (PLF) enables practitioners to monitor and analyze animal growth and health conditions, leading to improved productivity and welfare. However, the full potential of PLF remains elusive, as single-modality approaches are inherently limited by challenges such as environmental sensitivity and data sparsity. Multimodal fusion emerges as a critical pathway to overcome these limitations, yet the field currently lacks a systematic framework to synthesize its rapid developments. This work presents the first comprehensive review that specifically addresses this gap, offering a novel taxonomy to categorize multimodal fusion in PLF studies from 2019 to 2025. We introduce a distinctive analytical framework based on homogeneous (e.g., vision-vision, wearable sensor-wearable sensor) and heterogeneous (e.g., sensor-acoustic-text) fusion paradigms, under which we meticulously analyze livestock species distributions and fusion strategies across 66 surveyed publications. Our analysis reveals that feature-level fusion dominates (73% of studies), while homogeneous fusion accounts for 68% of implementations, with wearable sensor-based approaches being particularly prevalent. The superior robustness of these fused methods is demonstrated across key applications spanning individual identification, morphometric analysis, behavior recognition, event detection, and disease diagnosis. Beyond technical analysis, we provide a curated collection of open-access multimodal datasets that serve as valuable benchmarks. We further propose strategic recommendations for fusion paradigm and strategy selection, and highlight persistent challenges such as modality heterogeneity and alignment, missing modalities, and computational complexity and efficiency, while outlining potential solutions and emerging research directions. By contextualizing technological advances within real-world deployment constraints, this review aims to establish a foundational reference and accelerate the development of practical, sustainable multimodal PLF systems that genuinely enhance animal welfare and productivity.
No takes yet. Share an insight, caveat, or question.
Mao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: