Species Prediction Guide¶

Learn how to use CulicidaeLab's AI-powered species identification system to accurately identify mosquito species from photographs.

Overview¶

The Species Prediction feature uses state-of-the-art machine learning models to identify mosquito species from uploaded images. The system provides confidence scores, alternative predictions, and detailed species information to help you make accurate identifications.

Getting Started¶

Step 1: Access the Prediction Page¶

Navigate to the Predict tab in the main navigation bar
You'll see the species prediction interface with an image upload area

Step 2: Prepare Your Image¶

For best results, ensure your mosquito image meets these criteria:

Image Quality Requirements: - Resolution: Minimum 224x224 pixels (higher resolution preferred) - Format: JPEG, PNG, or WebP - File Size: Maximum 10MB - Lighting: Good, even lighting without harsh shadows - Focus: Sharp focus on the mosquito specimen - Background: Clean, uncluttered background preferred

Specimen Positioning: - Dorsal View: Top-down view showing wing patterns and body markings - Lateral View: Side view showing leg patterns and body profile - Full Specimen: Complete mosquito visible in frame - Scale: Mosquito should fill a significant portion of the image

Step 3: Upload Your Image¶

Drag and Drop: Drag your image file directly onto the upload area
File Browser: Click "Choose File" to browse and select your image
Camera Capture: Use "Take Photo" to capture directly from your device camera (mobile/tablet)

The system will automatically process your image once uploaded.

Understanding Prediction Results¶

Confidence Scores¶

Each prediction includes a confidence percentage:

90-100%: Very high confidence - likely accurate identification
70-89%: High confidence - good identification with minor uncertainty
50-69%: Moderate confidence - consider alternative predictions
Below 50%: Low confidence - manual verification recommended

Result Components¶

Primary Prediction: - Species name (scientific and common names) - Confidence percentage - Thumbnail of reference image

Alternative Predictions: - Up to 5 alternative species suggestions - Ranked by confidence score - Useful for verification and comparison

Species Information: - Quick facts about the identified species - Geographic distribution - Medical importance - Link to detailed species profile

Advanced Features¶

Model Selection¶

Choose from different AI models based on your needs:

Classification Models: - EfficientNet-B4: Best overall accuracy for general identification - ResNet-50: Fast processing with good accuracy - Vision Transformer: Excellent for complex specimens

Detection Models: - YOLOv8: Locates mosquitoes in complex images - Faster R-CNN: High accuracy detection with bounding boxes

Segmentation Models: - Mask R-CNN: Precise outline detection - U-Net: Detailed specimen segmentation

Batch Processing¶

Process multiple images simultaneously:

Select "Batch Upload" mode
Upload up to 20 images at once
Review results in a grid layout
Export results as CSV or JSON

API Integration¶

For programmatic access, use the prediction API:

import requests

# Upload image for prediction
url = "http://localhost:8000/api/v1/predict"
files = {"file": open("mosquito.jpg", "rb")}
response = requests.post(url, files=files)
result = response.json()

print(f"Species: {result['species']}")
print(f"Confidence: {result['confidence']:.2%}")

Step-by-Step Tutorial: Identifying an Aedes aegypti¶

Let's walk through a complete identification process:

Step 1: Image Preparation¶

Scenario: You have a mosquito specimen collected during field work and need to identify the species.

Image Setup: - Place specimen on white background - Use macro lens or close-up mode - Ensure dorsal view is clearly visible - Check that wing patterns and leg markings are sharp

Step 2: Upload and Initial Prediction¶

Navigate to the Predict page
Upload your prepared image
Wait for processing (typically 2-3 seconds)
Review the initial prediction results

Expected Results: - Primary prediction: Aedes aegypti (85% confidence) - Alternative: Aedes albopictus (12% confidence) - Other alternatives with lower confidence

Step 3: Verify the Identification¶

Key Features to Check: - Lyre-shaped markings: White scales forming lyre pattern on thorax - Leg banding: White bands on legs, especially hind legs - Wing scales: Dark scales with white patches - Size: Medium-sized mosquito (4-7mm)

Cross-Reference: 1. Click on the species name to view detailed profile 2. Compare your specimen with reference images 3. Check geographic distribution - is Aedes aegypti found in your area? 4. Review morphological characteristics

Step 4: Confirm or Adjust¶

If Confident in Identification: - Record the species identification - Note the confidence score for your records - Save or export the results

If Uncertain: - Try uploading additional angles of the same specimen - Compare with alternative predictions - Consult expert resources or seek professional verification - Consider environmental context (habitat, season, location)

Troubleshooting Common Issues¶

Low Confidence Scores¶

Problem: All predictions show confidence below 50%

Solutions: 1. Improve Image Quality: - Use better lighting - Increase image resolution - Ensure sharp focus - Clean specimen if possible

Try Different Angles:
Upload dorsal (top) view
Include lateral (side) view
Capture close-up of diagnostic features
Check Specimen Condition:
Damaged specimens may be harder to identify
Missing parts (legs, antennae) affect accuracy
Consider if specimen is within model training data

Unexpected Results¶

Problem: Prediction doesn't match expected species

Troubleshooting Steps: 1. Verify Image Quality: Ensure specimen is clearly visible 2. Check Geographic Range: Is predicted species found in your area? 3. Review Alternatives: Look at other high-confidence predictions 4. Consider Morphological Variation: Some species show significant variation 5. Seek Expert Opinion: Consult with entomologists for difficult cases

Technical Issues¶

Problem: Upload fails or processing errors

Solutions: 1. File Format: Ensure image is in supported format (JPEG, PNG, WebP) 2. File Size: Reduce file size if over 10MB 3. Internet Connection: Check network connectivity 4. Browser Compatibility: Try different browser or clear cache 5. Server Status: Check if service is temporarily unavailable

Best Practices¶

Field Collection¶

Documentation: - Record GPS coordinates of collection site - Note date, time, and weather conditions - Document habitat type and breeding sites - Take multiple photos from different angles

Specimen Handling: - Preserve specimens properly for photography - Use appropriate mounting techniques - Avoid damage to diagnostic features - Store specimens in suitable conditions

Photography Tips¶

Equipment: - Use macro lens or close-up filters - Employ ring flash or diffused lighting - Use tripod for stability - Consider focus stacking for depth of field

Technique: - Fill frame with specimen - Ensure even lighting - Capture multiple angles - Include scale reference when possible

Data Management¶

Record Keeping: - Save prediction results with metadata - Link predictions to collection records - Track confidence scores over time - Note any manual verifications

Quality Control: - Cross-reference with field guides - Seek expert verification for important identifications - Maintain database of verified specimens - Regular calibration with known species

Integration with Other Features¶

Map Visualization¶

Link your predictions to geographic data:

After species identification, click "Add to Map"
Enter collection coordinates
View your observation on the interactive map
Contribute to community surveillance data

Species Database¶

Explore detailed species information:

Click species name in prediction results
Access comprehensive species profile
View distribution maps and habitat information
Learn about medical importance and control measures

Disease Information¶

Understand vector potential:

Check if identified species is a disease vector
Access disease profiles for relevant pathogens
Review prevention and control strategies
Understand public health implications

Frequently Asked Questions¶

General Questions¶

Q: How accurate are the AI predictions? A: Our models achieve >90% accuracy on test datasets, but real-world performance varies based on image quality and specimen condition. Always consider confidence scores and verify important identifications.

Q: Can I identify larvae or pupae? A: Currently, the system is optimized for adult mosquitoes. Larval and pupal identification requires different approaches and is not yet supported.

Q: What species are included in the database? A: The system can identify 46 mosquito species commonly found in research and surveillance contexts. The database focuses on medically important species and common pest species.

Technical Questions¶

Q: Can I use the system offline? A: The web interface requires internet connectivity. However, the models can be deployed locally for offline use with appropriate technical setup.

Q: Is there a mobile app? A: The web interface is mobile-responsive and works well on smartphones and tablets. A dedicated mobile app is under consideration for future development.

Q: Can I contribute training data? A: Yes! We welcome high-quality, verified specimens to improve model performance. Contact the development team for contribution guidelines.

Data and Privacy¶

Q: What happens to my uploaded images? A: Images are processed for identification but not permanently stored unless you explicitly choose to contribute them to the database. See our privacy policy for details.

Q: Can I access my prediction history? A: Currently, predictions are not stored long-term. We recommend saving important results locally. User accounts with history tracking are planned for future releases.

Q: Is the service free to use? A: Yes, CulicidaeLab is open source and free for research, education, and public health applications.

Getting Help¶

Support Resources¶

Documentation: Comprehensive guides and API documentation
GitHub Issues: Report bugs and request features
Community Forum: Ask questions and share experiences
Email Support: Direct contact for specific issues

Expert Consultation¶

For challenging identifications or research applications:

Academic Partnerships: Collaborate with entomology departments
Professional Networks: Connect with medical entomologists
Taxonomic Experts: Consult with mosquito systematists
Public Health Agencies: Work with vector control professionals

Training and Workshops¶

Online Tutorials: Video guides for common workflows
Webinars: Regular training sessions for new users
Workshops: In-person training for research groups
Certification: Proficiency certification for professional use

Next Steps¶

After mastering species prediction:

Explore Map Visualization: Learn to visualize and analyze geographic patterns
Study Disease Information: Understand vector-pathogen relationships
Contribute Data: Add your observations to the community database
Advanced Analysis: Use API for custom applications and research workflows

For technical users: - API Documentation: Complete reference for programmatic access - Model Details: Technical specifications and performance metrics - Integration Guides: Connect with existing research workflows - Development: Contribute to open source development