Abstract / Description of output
Timely information about the necessity of thinning in the forest is vital for forest management to maintain a healthy forest while maximizing income. Currently, very high spatial resolution remote sensing data can provide crucial assistance for the experts to evaluate the maturity of thinnings. Yet, this task is still predominantly determined in the field and demands extensive resources. This paper presents a deep convolutional neural network (DCNN) to detect the necessity and urgency of thinnings by using only remote sensing data. The approach uses very high spatial resolution RGB and near-infrared orthophotos, canopy height model (CHM), digital terrain model (DTM), slope and the reference data, in this case from spruce dominated forests in the Austrian Alps. After tuning, the model achieves a test set F1 score of 82.23%, which is practically usable. We conclude that DCNNs are capable of detecting the need for thinning in forests. In contrast, attempts of assessing the urgency of thinnings with DCNNs proved to be unsuccessful. However, additional data such as age or yield class has the potential of improving the results. Investigation of the influence of the individual input features showed that orthophotos appear to contain the most relevant information for detecting the demand for thinning. Moreover, we observe a gain in performance by adding CHM and slope, whereas adding the DTM harms the model’s performance.
Original language | English |
---|---|
Article number | 419 |
Pages (from-to) | 1-24 |
Number of pages | 24 |
Journal | Algorithms |
Volume | 16 |
Issue number | 9 |
DOIs | |
Publication status | Published - 1 Sept 2023 |
Keywords / Materials (for Non-textual outputs)
- forest management
- agroforestry
- computer vision
- deep learning
- thinning