PawSense: A Computer Vision-Based Cat Breed Recognition and Veterinary Decision Support System

PawSense is an open-source project combining computer vision and AI to automatically recognize cat breeds and provide veterinary decision support. Developed by MinhNhatTyson and hosted on GitHub (released 2026-06-01), it aims to innovate pet healthcare by offering practical tools for pet owners and professionals. Its core functions include:

1. Cat breed identification (auto recognition, multi-breed support, confidence scores, mixed breed detection)
2. Veterinary decision support (breed-specific health profiles, risk assessments, care advice, symptom-disease associations)

The global pet market is growing, and cats are among the most popular pets. Different cat breeds have distinct health traits and disease susceptibilities:

• Persian cats: Prone to polycystic kidney disease and respiratory issues
• Maine Coons: Higher risk of cardiomyopathy
• Siamese cats: Susceptible to dental and respiratory problems
• Ragdolls: May carry hypertrophic cardiomyopathy genes

Accurate breed identification is key to preventive care. PawSense addresses this need by leveraging AI to provide intelligent breed recognition and health assessment solutions.

Breed Recognition System

• Features: Auto identify breeds from photos, support mainstream breeds (Persian, British Shorthair, American Shorthair, Siamese, Ragdoll, Maine Coon), provide confidence scores, detect mixed breeds.
• Tech: Convolutional Neural Networks (CNN) using pre-trained models (ResNet, EfficientNet) with transfer learning; data augmentation (rotation, scaling, flipping); model optimization (quantization, pruning) for edge deployment (mobile devices)

Veterinary Decision Support System

• Features: Display breed health profiles, assess disease risks, offer targeted care advice (diet, exercise, medical), link observed symptoms to breed-specific diseases.
• Knowledge Base: Integrate veterinary literature and breed health databases; build breed-disease association maps; update with latest medical research.

Computer Vision

• Frameworks: PyTorch/TensorFlow
• Image processing: OpenCV
• Detection modules: YOLO/Faster R-CNN (optional)
• Cross-platform deployment: ONNX

Backend

• API: Python/Flask or FastAPI
• Data storage: PostgreSQL/MongoDB
• Caching: Redis
• Asynchronous tasks: Celery

Frontend

• Web: React/Vue.js
• Mobile: React Native/Flutter (optional)
• Camera integration: Direct photo upload support

MLOps

• Experiment tracking: MLflow
• Containerization: Docker
• CI/CD: Automated testing and deployment pipelines

For Pet Owners

• Adoption helper: Identify shelter cat breeds
• Health monitoring: Understand breed-specific risks
• Daily care: Get personalized feeding and care advice

For Vet Clinics

• Initial diagnosis aid: Quick access to breed health backgrounds
• Diagnostic reference: Symptom-breed disease association hints
• Client education: Explain breed-specific issues to owners

For Animal Shelters

• Fast classification: Auto identify shelter cat breeds
• Matching recommendations: Suggest suitable adoptive families based on breed traits
• Medical priority: Identify breeds needing special care

1. Breed Similarity: Many breeds look alike (e.g., British Shorthair vs Russian Blue) Solution: Use fine-grained classification, focus on unique details (ear shape, eye color, fur texture), integrate multi-angle images

2. Mixed Breed Recognition: Most cats are mixed Solution: Adopt multi-label classification, output breed proportion (e.g., "70% Persian +30% British Shorthair"), add "mixed" as a category

3. Image Quality Issues: Photos may have poor lighting, bad angles, or cluttered backgrounds Solution: Image preprocessing pipeline (denoising, enhancement, standardization); cat detection and cropping; UI prompts to guide users to take high-quality photos

4. Data Privacy & Security: Pet images may contain EXIF location data; health data is private Solution: Local processing option (no cloud upload); encrypted storage; GDPR compliance; transparent data usage policy

Training Data

• Public datasets: Oxford-IIIT Pet Dataset
• Web crawling: Compliant collection of breed-labeled images
• Crowdsourced annotation: Vet experts verify labels
• Data balancing: Address rare breed sample shortages

Performance Metrics

• Top-1 accuracy (main breeds)
• Top-5 accuracy (correct breed in top 5 candidates)
• Inference speed: Target <100ms per image
• Model size: Optimized for mobile deployment

Ethical Considerations

• Avoid breed discrimination (no bias against non-purebred cats)
• Clarify system advice vs professional vet diagnosis
• Ensure training data covers diverse regions/environments
• Prevent over-medication based on low-risk predictions

Conclusion

PawSense demonstrates AI's potential in pet healthcare. Future extensions include multi-species support (dogs, rabbits), health detection (skin diseases, eye issues), age/emotion recognition, AR integration. The project welcomes community contributions (data, model improvements, knowledge expansion, localization, feedback)