Aerial image plant species detection refers to the automated identification and mapping of plant species from imagery captured by aircraft, satellites, or UAVs, using computer vision and machine learning techniques to classify vegetation at individual plant, crown, or patch level across large spatial extents. The sensor inputs employed range from standard RGB cameras and multispectral sensors (red, green, blue, red-edge, near-infrared bands) through to hyperspectral imagers capturing hundreds of contiguous spectral bands, each offering different trade-offs between species discrimination power, cost, and data volume. Deep learning approaches — particularly convolutional neural networks trained on annotated aerial image datasets — have become the dominant analytical paradigm, with models capable of distinguishing tree species, identifying invasive weed species, detecting disease or stress symptoms, and mapping understorey vegetation composition from canopy-penetrating imagery. Object detection frameworks (YOLO variants, EfficientDet) and semantic or instance segmentation models (DeepLab, SegFormer, Mask2Former) are widely applied, with model performance heavily dependent on the quality and representativeness of training data. In Australian forestry and land management, aerial plant species detection has been applied across several critical domains: mapping invasive weeds such as lantana, blackberry, and radiata pine wildings in native ecosystems; identifying threatened plant species in harvest planning zones; monitoring revegetation success in post-harvest or restoration areas; and distinguishing eucalypt species assemblages to support biodiversity offset accounting. UAV-based surveys offer the highest spatial resolution and operational flexibility for local assessments, while satellite imagery (increasingly from commercial providers offering sub-metre resolution) enables landscape-scale species distribution mapping. Integrating aerial species detection outputs with existing vegetation mapping frameworks such as NVIS (National Vegetation Information System) is an active challenge, as is developing the large, geographically diverse labelled datasets needed to train robust models across Australia's exceptionally diverse and spatially variable flora.
| Field |
Value |
| Company |
Dendra Systems |
| Country |
UK |
| Type |
Company |
| Employees |
100 |
| FWPA RD&E |
7.3 |