Migrating Monarch Butterfly Localization UsingMulti-Modal Sensor Fusion Neural Networks

Generate training data

The training data for temeprature is included as ./dataset/Temp_train_16.mat and ./dataset/Temp_valid_16.mat

To generate the training data for light, you need to run the MATLAB script Generate_trainset_light.m. Then, you are expected to get ./dataset/Light_train_8.mat and ./dataset/Light_valid_8.mat.

Update: for the privacy consideration, we removed the raw data from this repository. Instead, we share the processeed training set following this link

Train the neural networks

Simply run train_light.py and train_temp.py. The logs will be stored in ./logs and the trained models will be stored in ./model

Testing the neural networks

The test data for temperature and light are included in ./testdata/Test_set_temp and ./testdata/Test_set_light. The light and temperature curves are sampled around the rounded ground truth (for the seek of privacy). We provide 20 test examples.

The pretrained models are included in ./model and you can directly run this part without re-training the neural networks

Generate heatmaps

Run test_light.py and test_temp.py, and the heatmaps (confidence) will be stored in ./results

Visualization

Run visualization.m. The visualization of heatmaps are stored in ./results/heatmap_visual.

Here is an example of visualization plot

Here are the MLE localization results over all test data

Reference

Mingyu Yang, Roger Hsiao, Gordy Carichner, Katherine Ernst, Jaechan Lim, Delbert A. Green II, Inhee Lee, David Blaauw, and Hun-Seok Kim, "Migrating Monarch Butterfly Localization UsingMulti-Modal Sensor Fusion Neural Networks", EUSIPCO 2020