Multimodal Cattle Behavior Recognition
Wearable and farm-sensing based cattle behavior recognition using time-series sensor data at Qatar University.
Ongoing research at Qatar University under Dr. Amith Khandakar on multimodal cattle behavior recognition using wearable and farm-sensing modalities. The project targets automatic recognition of cattle behaviors such as feeding, locomotion, standing, and lying using time-series sensor data collected from real farm environments.
Core Research Challenge:
Real farm deployments do not always provide complete or consistent sensor observations. Sensors may fail, modalities may be unavailable, and models trained on one group of animals may not generalize to unseen animals or farm conditions. The research therefore investigates how multimodal time-series information can be effectively fused during training while retaining the possibility of deployment under limited sensor configurations.
Research Focus:
- Multimodal fusion of time-series sensor streams from wearable and farm-sensing devices
- Cross-animal generalization and domain adaptation
- Missing-modality robustness for real-world deployment
- Interpretability and explainability of behavior recognition decisions
- Reproducible experimentation and statistical evaluation
My Contributions:
- Model development and multimodal architecture design
- Time-series preprocessing and feature extraction
- Cross-animal evaluation and statistical analysis
- Interpretability analysis of model predictions
- Manuscript preparation
Affiliation: Qatar University, Department of Electrical Engineering
Status: Model development in progress