Abstract
Accurate and timely monitoring is imperative to the resilience of forests for economic growth and climate regulation. In the UK, forest management depends on citizen science to perform tedious and time-consuming forest-related data collection tasks. In this study, UAV allowed aerial surveying to observe Forest Health Indicators (FHI) inaccessible from the ground. However, many FHI such as burrows and deadwood around the trunk can only be observed from the ground under the tree canopy. Hence, in addition to a UAV, a quadruped robot was brought into play to observe these FHI from the ground to reduce the time and effort of forest personnel to carry out various surveying tasks. Moreover, the forest monitoring is required to be based on algorithms that consume less computation power to conserve the battery. Therefore, machine learning algorithms are analysed in terms of their size to be used by the external sensory platform equipped with sensors, computing and communication modules. Custom datasets are constructed to train the object detection algorithm YOLOv5 for recognising FHI, fire, and persons. Our results reveal that YOLOv5s is lightweight and easy to train for FHI detection while performing close to real-time, cost-effective and autonomous forest monitoring.
Original language | English |
---|---|
Type | Dataset |
Media of output | Image dataset of forest health indicators |
Publisher | MDPI |
Publication status | Published (VoR) - 6 Jun 2022 |
Keywords
- Forest health indicators; object detection; YOLOv5
- WebRTC
- 5G
- Real-time monitoring