Marine Life Identification

Instructions

Build AI-powered marine life identification systems that accept photos or text descriptions and return species identification with confidence scores, habitat context, behavioral information, conservation status, and regional distribution.

1. Identification Workflow

Process identification requests through this pipeline:

1. Input classification: Determine input type (photo, text description, or both)
2. Feature extraction: Identify distinguishing characteristics (shape, color, pattern, size, behavior, habitat)
3. Candidate matching: Produce ranked list of species matches with confidence scores
4. Context enrichment: Add habitat, behavior, conservation, and distribution data
5. Result presentation: Return structured identification with supporting evidence

2. Species Database Schema

Structure the marine life database for efficient identification:

species
├── id (slug: "blue-tang")
├── common_names: ["Blue Tang", "Regal Tang", "Palette Surgeonfish"]
├── scientific_name: "Paracanthurus hepatus"
├── taxonomy: { family, order, class, phylum }
├── category: "fish" | "coral" | "invertebrate" | "marine_mammal" | "sea_turtle" | "ray_shark"
├── identification
│   ├── key_features: ["Bright blue body", "Black palette pattern", "Yellow tail"]
│   ├── size_range: { min_cm, max_cm, typical_cm }
│   ├── similar_species: [{ species_id, distinguishing_feature }]
│   └── color_variants: [{ name, description, image_ref }]
├── habitat
│   ├── depth_range: { min_m, max_m, typical_m }
│   ├── environment: ["coral reef", "lagoon", "outer reef slope"]
│   ├── substrate: ["live coral", "rubble"]
│   └── behavior: ["schooling", "grazing algae", "sheltering in coral"]
├── distribution
│   ├── regions: ["Indo-Pacific", "Red Sea", "Great Barrier Reef"]
│   ├── countries: ["Australia", "Indonesia", "Philippines", ...]
│   └── dive_sites: [{ name, region, likelihood: "common" | "occasional" | "rare" }]
├── conservation
│   ├── iucn_status: "LC" | "NT" | "VU" | "EN" | "CR" | "DD"
│   ├── population_trend: "stable" | "declining" | "increasing" | "unknown"
│   └── threats: ["habitat loss", "collection for aquarium trade"]
└── diver_notes
    ├── best_time_to_see: "Morning feeding hours"
    ├── photography_tips: "Use wide angle on schools; macro for juvenile details"
    └── interaction_guidelines: "Observe from distance; do not touch or chase"

3. Photo Identification Process

When processing a photo for identification:

• Primary features: Body shape, fin configuration, coloration pattern, size estimate
• Secondary features: Eye shape, mouth type, scale pattern, markings
• Context clues: Background habitat (reef, sand, pelagic), depth indicators, associated species
• Confidence scoring:
• 90-100%: High confidence, distinctive species with clear features visible
• 70-89%: Probable match, most features align but some ambiguity
• 50-69%: Possible match, present top 2-3 candidates with distinguishing features
• Below 50%: Insufficient for identification, request additional photos or description

4. Text Description Identification

When processing a text description:

• Extract structured features from natural language: color, size, shape, location, behavior, depth
• Handle imprecise descriptions (“medium-sized blue fish”) by expanding to candidate range
• Ask clarifying questions when description matches 5+ species equally
• Weight recent observation location heavily — many species are region-specific

5. Confidence Scoring Model

Score identifications on these axes:

Factor	Weight	Description
Visual match	40%	How closely features match the species profile
Regional match	25%	Whether the species is found in the observation region
Habitat match	15%	Whether depth, substrate, and environment align
Behavioral match	10%	Whether observed behavior matches known patterns
Size match	10%	Whether reported size falls within species range

6. Conservation Context

Always include conservation information in identification results:

• IUCN Red List status with plain-language explanation
• Population trend and primary threats
• Protected status in the observation region (if applicable)
• Responsible interaction guidelines (no touching, distance requirements, flash photography rules)
• Report channels for rare or unusual sightings

7. Regional Distribution Integration

Connect identifications to dive planning:

• List regions and specific dive sites where the species is commonly seen
• Seasonal availability (migrations, spawning aggregations, whale shark seasons)
• Best conditions for observation (time of day, tide, current)
• Connect to the dive-planning skill for trip recommendations

Inputs Required

• Photo of marine life (if available) or text description
• Observation location (region, country, or specific dive site)
• Depth of observation (approximate)
• Date and time of observation
• Any behavioral observations (feeding, schooling, cleaning, mating)

Output Format

• Species identification with confidence score
• Common and scientific names
• Key identifying features that match the input
• Habitat and behavioral context
• Conservation status (IUCN)
• Regional distribution and where to see this species
• Similar species to differentiate
• Photography and interaction guidelines

Anti-Patterns

• Single-answer identification at low confidence: Always present multiple candidates when confidence is below 70%.
• Ignoring regional context: A species identification that doesn’t occur in the observation region is likely wrong regardless of visual similarity.
• Omitting conservation status: Every identification is a teaching moment about marine conservation.
• Over-relying on color alone: Many marine species change color by mood, depth, age, and sex. Use body shape and pattern as primary identifiers.
• No similar species comparison: Users need to know what else it could be, especially for look-alike species with different conservation statuses.
• Flat database without relationships: Species exist in ecosystems. Link to habitat, cleaning station relationships, predator-prey context.