Published December 1, 2021
| Version v1
Software
Open
Mapping wheel-ruts from timber harvesting operations using deep learning techniques in drone imagery - codes
Creators
- 1. Norwegian Institute of Bioeconomy Research NIBIO
- 2. Stellenbosch University, South Africa
Description
- CNN based automatic wheel rut detection in harvested forest areas. The published paper and more details on the study can be found at: https://doi.org/10.1093/forestry/cpac023
- The codes with detailed explanation are available at: https://github.com/SmartForest-no/wheelRuts_semanticSegmentation - Please refer it.
- The model is saved as 'weights' and is a .json file - used for predicting Wheel Ruts and Non-Wheel Ruts. The weights can be obtained from the NEW VERSION of doi.
- The model was trained for multiple epochs with ~2500 images from 20 harvested forest sites in Norway. The model is ResNet50 + UNet semantic segmentation model with pre-trained weights from ImageNet to train ResNet50 model. UNet has been trained from scratch. For optimal identification of the wheel ruts, we recommend using 20 meters x 20 meters of images.
Files
codes.zip
Files
(100.0 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:03f242b63d5cf2a34670399098e84cf7
|
100.0 kB | Preview Download |
Additional details
Related works
- Is supplemented by
- Preprint: https://zenodo.org/record/5208232#.YS3tko4zZaQ (URL)