The TRN system–that has been called “FederNet”– that combines the strength of a convolutional neural network (CNN or convnet) with the robustness of projective invariants theory. FederNet is specifically designed for a lunar mission but its applicability could be furthed extended to other airless bodies. The core algorithm is the crater detection algortihm (CDA), an implementation of a version of Mask R-CNN(modified by akTwelve) on Python 3, Keras, and TensorFlow 2 for my master thesis. The model generates bounding boxes and segmentation masks for each instance of craters in the image. It's based on Feature Pyramid Network (FPN) and a ResNet101 backbone.
The repository includes:
- Source code of crater detection algorithm (CDA)
- Source code of crater matching algorithm (CMA)
- Source code of position estimation algorithm (PEA)
- Implementation of the EKF, through filterpy library
- Pre-trained weights for Lunar DEM LOLA-KAGUYA
- Datasets of training, test and validation (instance segmentation)
Use this bibtex to cite this repository:
@misc{FederNet_TRN_2020,
title={FederNet: a robust convnet-based terrain relative navigation system for planetary applications},
author={Roberto Del Prete},
year={2020},
publisher={Github},
journal={GitHub repository},
howpublished={\url{https://github.com/SirBastiano/Mask_RCNN}},
}