🪲 Beetle Drought Predictor (HDR-SMood Challenge)
This repository hosts the official model weights for Team nebiyu's first-place submission to the Imageomics HDR-SMood Challenge.
Model Description
The model is a deep learning architecture designed to predict environmental drought indices (Standardized Precipitation Evapotranspiration Index - SPEI) across three continuous timeframes (30-day, 1-year, and 2-year). It achieves this by performing multimodal feature extraction, combining robust image processing of anatomical beetle specimens with structured ecological metadata.
Developed by: Nebiyeleul Yifru (University of Maryland, Baltimore County - UMBC)
Model Type: Deep Learning Regression (Vision + Metadata)
Format: Keras (.keras)
🔗 Training Code & GitHub Repository
To comply with FAIR data standards, the complete end-to-end pipeline—including data cleaning, model training, and performance evaluation (scatterplots/metrics)—is publicly available on GitHub.
GitHub Repository: cable13/Beetle_Drought_Prediction
Training Notebook: model_training_and_evaluation.ipynb
Inference Code: model.py
How to Use
To load this model locally for inference:
import tensorflow as tf
Load the model
model = tf.keras.models.load_model("best_beetle_model.keras")
The model expects a tuple of inputs: (images, metadata)
predictions = model.predict([image_batch, metadata_batch])
For the complete batch inference pipeline, including dynamic image resizing and metadata handling, please refer to the model.py script in the linked GitHub repository.
