Skip to content

Convolutional Neural Networks to predict forest stand variables with help of voxelized ALS data

Notifications You must be signed in to change notification settings

balazsan/ALS_NNs

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ALS_NNs

Comparison of k-Nearest Neighbors (k-NN) method and Neural Networks (NN) to predict forest stand variables with help of ALS data


sample_plot _data:

CSV files containing mean diameter in cm (d), mean height in m (h) and total growing stock in m3/ha (v).


features:

CSV files with features calculated using ALS_feature_calc.R. Note, no augmentation done here as it would produce the same values for most of the features.


voxel_data:

voxel_data.npz: 4D-numpy arrays with voxel data for 2D and 3D-CNN. Numeric values represent the number of points within each voxel. Pre-augmented training data is used in 2D-CNN, augmentation carried out on-the-fly in 3D-CNN. The compressed file contains all voxel datasets (training (augmented), validation, test).


k-NN:

lidR_lasfeat.R: Functions to calculate features using ALS point cloud files.

ALS_feature_calc.R: Wrapper script to call lidR_lasfeat.R. Showing the use of the relevant functions, no real data provided due to data distribution restrictions.

ALS_feature_sel.R: Feature selection and feature weight search.

knn_funcs_group_28052019.R: k-NN and genalg functions.


ANN (Artificial Neural Network):

ANN.R: Predicting forest stand variables using simple ANN based on ALS features.

keras_tf_funcs.R: Functions to estimate forest attributes based on field measurements and remote sensing features using Keras with TensorFlow (TF) backend.


2D-CNN:

2D-CNN_AlexNet.py: AlexNet-type 2D CNN for forest variable prediction. Input 4D numpy array (voxelized ALS data) and csv table with response variables (field data, e.g. volume, mean height, diameter, etc.)


3D-CNN:

InceptionV3_multi_sepmod_autoaug.py: 3D-CNN based on InceptionV3 implemented by Elias Ayrey for forest variable prediction. Input 4D numpy array (voxelized ALS data) and csv table with response variables (field data, e.g. volume, mean height, diameter, etc.). This version has been edited from its original (https://github.com/eayrey/3D-Convolutional-Neural-Networks-with-LiDAR/tree/master/InceptionV3-3D_Neural_Network) to carry out multi-regression tasks. The possibility to do training data augmentation on-the-fly and to train the network using all response variable at the time or use alternate training has been added as well. In alternate training the network is trained using one response variable after the other.

About

Convolutional Neural Networks to predict forest stand variables with help of voxelized ALS data

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published