Coral Image Identifier (CoralID) is an end-to-end computer vision platform and scientific editorial interface built to automate coral species taxonomy (e.g., Acropora, Montipora, Zoanthids) from subsea photography. DEEP LEARNING MODEL DESIGN & CV PIPELINE: - Model Architecture & Transfer Learning: Upgraded backbone from MobileNetV2 (69% val acc) to EfficientNetV2-B0, boosting validation accuracy to 81%. Executed a two-stage training pipeline (Stage 1 warmup head + Stage 2 unfreezing top layers 200–270 for fine-tuning at a 1e-5 learning rate). - Hyperparameter Optimization: Utilized Keras-Tuner (Hyperband) to optimize dense head dropout rates and learning rate schedules. - LAB-Space CLAHE Enhancement: Implemented Contrast Limited Adaptive Histogram Equalization on the L (Lightness) channel in LAB color space to normalize uneven exposure and blue-spectrum underwater lighting while preserving true species coloration. - Global Perceptual Deduplication: Developed a 64-bit pHash algorithm to remove cross-class visual duplicates across dataset categories, preventing data leakage. - Class Imbalance & Evaluation: Applied Scikit-Learn mathematical class weighting to prevent minority class bias, evaluating performance via confusion matrices and precision/recall reporting. - Automated Ingestion: Built scrapers for retailer APIs, Bing Image Search, and integrated the iNaturalist community observation API. SCIENTIFIC UX DESIGN & DASHBOARD (CoralID v4.0): - Editorial Scientific Interface: Designed a high-fidelity dashboard built with React (Vite), Tailwind CSS, and Framer Motion, utilizing OKLCH color spaces and a tactile editorial grain aesthetic. - HUD Neural Ingestor: Created an interactive drag-and-drop upload zone with real-time scan-line animations and live technical telemetry. - Visual Architecture Pipeline: Implemented an interactive 4-stage validation infographic (Preprocessing, Feature Extraction, Neural Analysis, Taxonomy Output). - Local Research Archives: Engineered persistent local browser history saving specimen runs, timestamped telemetry, and species confidence distribution metrics. - FastAPI REST Backend: Served real-time model inference via FastAPI and Uvicorn with interactive Swagger API telemetry endpoints.