Detection and localization of indigenous economic grasses in depth-color close-range aerial images, using a novel trainingless data-to-decision approach (DIKD Hierarchy), for the Sharjah pastures project
Traditional open-field plant monitoring approaches are costly, labor-intensive, and risky. Under arid conditions, the risks are elevated. They are associated with fatigue, dehydration, animals' attacks, etc. Plant detection is a pillar of precision agriculture. It is required for various crop m...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Smart Agricultural Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772375525000309 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Traditional open-field plant monitoring approaches are costly, labor-intensive, and risky. Under arid conditions, the risks are elevated. They are associated with fatigue, dehydration, animals' attacks, etc. Plant detection is a pillar of precision agriculture. It is required for various crop management processes, including: autonomous plantation, irrigation, and harvesting. Researchers utilized off-the-shelf supervised learning modules for plant detection, by re-training these modules on powerful computers. This research, in contrast, aims at developing an unsupervised trainingless method that runs on resource-limited edge-computers. The outcome of this research is an Artificial Intelligence (AI) tool for the detection of two economic indigenous forage crops, namely: Cenchrus ciliaris and Pennisetum divisum for the pastures project in the Emirate of Sharjah, United Arab Emirates (UAE). The developed tool detects plants inflorescences in depth-color close-range aerial images of the plants. A novel Decision Hierarchy (DIKD) - data, information, knowledge, and decision - is proposed and showcased in this research. The proposed DIKD approach detected the target species with an average accuracy of 0.98, using novel blob features: blob regularity and blob strawness. Experimental results demonstrate the robustness of these features for target plants detection under open environment conditions. This work helps create georeferenced maps of plants distribution within the pastures. These maps enable sustainable management of the Sharjah pastures, for example, by aiding remote assessment of the pastures' carrying capacity, and performing autonomous aerial-irrigation based on individual tussock needs. Moreover, these species-aware maps aid planning rotational grazing, and enable habitat restoration using seeds harvested from the resting paddocks. |
|---|---|
| ISSN: | 2772-3755 |