Publications Archive - Tech4Animals

Prototype-based Explainable Deep Learning for Sex Classification of Mountain Gazelles in the Wild

annazam — Sun, 20 Sep 2026 17:26:03 +0000

Standard deep learning tools in automated wildlife monitoring often function as opaque “black boxes,” hindering the trust required for ecological management. To address this challenge, this study presents an explainable visual AI framework that classifies the sex of endangered Mountain Gazelles from unconstrained camera trap imagery. The approach extends prototype-based deep learning by adapting visual exemplars across multiple spatial scales, capturing coherent anatomical features rather than rigid image patches. The model achieves 75% classification accuracy while transparently grounding predictions in recognized morphological traits, including male horn structures and female neck profiles. This multi-size architecture demonstrates that transparent artificial intelligence can deliver biologically meaningful demographic monitoring without forfeiting predictive performance.

The post Prototype-based Explainable Deep Learning for Sex Classification of Mountain Gazelles in the Wild appeared first on Tech4Animals.

SHIC-XE: Viewpoint-Invariant Explainability via Dense 2D-3D Correspondences: an Application to Equine Pain Recognition

annazam — Fri, 10 Jul 2026 17:32:04 +0000

Automated recognition of animal pain often relies on two-dimensional saliency maps, which suffer from severe flickering and instability as subjects move across changing camera viewpoints. To overcome this limitation, this study presents an explainability framework that leverages dense 2D-3D correspondence mapping to project neural network attention from video frames onto a canonical three-dimensional surface prototype. Evaluated on horses across post-surgical, orthopedic, and acute pain conditions, the approach achieved video-level classification F1 scores ranging from 0.67 to 0.80. Crucially, the method provides quantitative metrics that map model focus to specific facial regions, establishing statistically significant correlations between machine attention and expert veterinary pain scores for key features like ears and cheek muscles.

The post SHIC-XE: Viewpoint-Invariant Explainability via Dense 2D-3D Correspondences: an Application to Equine Pain Recognition appeared first on Tech4Animals.

Analysis of Human-oriented Facial Signals of the Domestic Dog using Eye Tracker Technology

annazam — Wed, 13 May 2026 17:37:56 +0000

While reading canine facial expressions is vital for animal welfare, how head morphology affects human visual processing of dog faces remains under-researched. This study used eye-tracking technology to compare human visual attention when viewing domestic dogs with short-muzzled (brachycephalic) versus typical (normocephalic) facial structures. Participants allocated roughly 45% more visual attention to flat-faced dogs through repeated gazes rather than longer individual fixations, reflecting higher cognitive effort to decode compressed features. Although viewers fixated heavily on the ears of brachycephalic dogs, major discrepancies emerged between self-reported focus and objective gaze measurements. These findings demonstrate that altered facial structure complicates signal interpretation, highlighting important implications for veterinary practice and human-canine communication.

The post Analysis of Human-oriented Facial Signals of the Domestic Dog using Eye Tracker Technology appeared first on Tech4Animals.

Predicting Guide Dog Career Success using Machine Learning and Large Language Models

annazam — Fri, 24 Apr 2026 17:41:20 +0000

High attrition rates in guide dog training programs cause substantial financial costs and resource strains when unsuitable dogs are identified late in development. To address this challenge, this study developed a machine learning framework combining structured behavioral assessments with qualitative trainer notes processed by large language models. By converting unstructured trainer comments from nearly one thousand dogs into numerical sentiment indicators, the pipeline captured subtle behavioral nuances alongside standard checklist metrics. Incorporating these textual observations significantly improved predictive accuracy over behavioral scores alone while generating clear, trainer-validated explanations for predictions. Ultimately, this hybrid approach demonstrates how natural language processing can optimize candidate screening and improve working dog selection workflows

The post Predicting Guide Dog Career Success using Machine Learning and Large Language Models appeared first on Tech4Animals.

Supervised Neural Style Transfer as an Augmentation Technique for Facial Landmark Detection

annazam — Tue, 07 Apr 2026 13:20:48 +0000

A major challenge in training accurate facial landmark detection models for animals is the scarcity of diverse data, as classical image augmentations often disrupt the crucial spatial alignments required for this task. To address this gap, this study introduces a novel data augmentation technique called Supervised Neural Style Transfer (SNST). Focusing on cats, the researchers synthetically expanded their training dataset by transferring textural styles from top-performing examples onto cropped facial images, which perfectly preserved the original geometric structure of the faces. This specific approach effectively decouples appearance from shape, forcing the models to rely on robust structural features rather than surface-level textures. Ultimately, the study demonstrates that SNST significantly outperforms traditional augmentation methods, substantially reducing detection errors and improving the models’ overall robustness.

The post Supervised Neural Style Transfer as an Augmentation Technique for Facial Landmark Detection appeared first on Tech4Animals.

Close Encounters of the Cat Kind: The influence of context and sex on facial signaling proximity in domesticated cats (Felis silvestris catus)

annazam — Wed, 01 Apr 2026 12:42:59 +0000

Accurately measuring animal social bonds often relies on analyzing spatial proximity, yet advanced automated tracking methods have rarely been applied to feline interactions. To address this gap, this study utilized AI-based computer vision systems to extract precise distance data from video recordings of domestic cats engaging in social facial signaling. Contrary to initial predictions, the analysis revealed that cats maintained significantly closer physical proximity during non-affiliative encounters than during friendly, affiliative interactions. Additionally, the sex of the interacting pairs strongly influenced their spacing, with female-female dyads staying the closest across all contexts. Ultimately, this work demonstrates how machine learning tools can uncover nuanced behavioral strategies that animals use to navigate social environments.

The post Close Encounters of the Cat Kind: The influence of context and sex on facial signaling proximity in domesticated cats (Felis silvestris catus) appeared first on Tech4Animals.

From Data to Dialogue: A Multi-Agent Chatbot Interface for Decision Support in Guide Dog Training

annazam — Sun, 22 Mar 2026 14:17:05 +0000

Guide dog training programs collect extensive behavioral and performance data, but trainers often struggle to utilize this information because it is locked in complex databases requiring technical expertise. To bridge this gap, this study introduces a multi-agent conversational chatbot that helps trainers easily access and interpret data about guide dogs. Using natural language queries, handlers can connect to a machine learning pipeline to compare historical records and predict training outcomes. An evaluation with professional trainers demonstrated that this dialogue-based interface made the underlying behavioral information highly approachable and interpretable. Ultimately, the system transforms scattered technical metrics into actionable insights, successfully empowering trainers to make practical, data-driven decisions.

The post From Data to Dialogue: A Multi-Agent Chatbot Interface for Decision Support in Guide Dog Training appeared first on Tech4Animals.

Comparing the Performance of Deep Learning Video-based Models and Trained Veterinarians in Cattle Pain Assessment

annazam — Wed, 18 Mar 2026 13:30:54 +0000

Accurately assessing pain in cattle remains a significant challenge because these prey animals naturally conceal signs of discomfort, which can lead to subjective and highly variable human evaluations. To address this limitation, this study introduced a novel, video-based deep learning framework designed to automatically detect subtle pain-related expressions and behavioral changes. Using computer vision techniques to analyze temporal and spatial features in video recordings of bulls, the system’s performance was compared against standardized scoring by trained veterinarians. The machine learning model achieved an impressive 97% accuracy in pain classification. This performance successfully outperformed video-based human assessments and matched the accuracy of real-time expert evaluations, offering a highly objective and scalable tool for monitoring livestock welfare.

The post Comparing the Performance of Deep Learning Video-based Models and Trained Veterinarians in Cattle Pain Assessment appeared first on Tech4Animals.

Canine Gait Analysis using Inertial Sensors and Deep Learning for Orthopedic and Neurological Disorders

annazam — Wed, 18 Mar 2026 13:26:29 +0000

Distinguishing between orthopedic and neurological gait abnormalities in dogs is notoriously difficult, as subtle movement changes often elude even experienced clinicians during standard visual inspections. To address this diagnostic challenge, this study equipped 29 dogs with wearable inertial sensors to capture precise movement data while they walked and trotted. Researchers then developed a lightweight deep learning model to automatically analyze the raw accelerometer and gyroscope signals. The model proved highly effective, achieving up to 96% accuracy in classifying the subjects as healthy, orthopedic, or neurological. Crucially, the research demonstrated that a single sensor mounted comfortably on the dog’s collar provided optimal classification performance, establishing an accessible and objective diagnostic aid for veterinary clinics.

The post Canine Gait Analysis using Inertial Sensors and Deep Learning for Orthopedic and Neurological Disorders appeared first on Tech4Animals.

Time Budget Analysis with Large Vision Language Models for Automated Pain Assessment in Horses

annazam — Fri, 06 Mar 2026 14:22:28 +0000

Assessing pain in horses is challenging because they instinctively hide discomfort, but shifts in their daily behavioral routines can provide vital clues. To bypass the impracticality of continuous manual observation, this study developed an automated video analysis pipeline using Large Vision Language Models (LVLMs). Researchers applied these models to stall surveillance videos of hospitalized horses to track time spent on activities like resting, feeding, and moving. These time budgets were then fed into machine learning classifiers to distinguish between painful and non-painful states. The system achieved an 80% accuracy in pain detection, proving that LVLMs offer a scalable, non-invasive solution for equine welfare monitoring without requiring expensive manual data annotation.

The post Time Budget Analysis with Large Vision Language Models for Automated Pain Assessment in Horses appeared first on Tech4Animals.