Oilpalm-RTMDet: An lightweight oil palm detector base on RTMDet
Oil palm, as an important economic crop, plays a significant role in yield estimation and plantation management, making accurate identification and counting crucial. However, due to the varying sizes of oil palm crowns, diverse tree shapes, and the frequent overlap of adjacent crowns, existing algor...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954125000093 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Oil palm, as an important economic crop, plays a significant role in yield estimation and plantation management, making accurate identification and counting crucial. However, due to the varying sizes of oil palm crowns, diverse tree shapes, and the frequent overlap of adjacent crowns, existing algorithms often struggle to achieve precise individual tree recognition and accurate counting. To address this challenge, this study proposes the Oilpalm-RTMDET model, which is based on the RTMDET algorithm. First, the SimSPPF structure is used to replace the SPPF structure of backbone, then the CSP-ELAN structure is used to replace the CSPLayer structure in the original neck, and the UpsamplingBiliearn is used as the upsampling method. Finally, the SGEAttention mechanism is introduced to enhance the extraction ability of oil palm tree features. The experimental results show that Oilpalm-RTMDET model has higher target detection accuracy, with AP50 and AP75 achieving 94.1 % and 58 % respectively, and FPS of 48.4 img/s, which is better than that of RTMDET model. Moreover, Oilpalm-RTMDET model can realize accurate and rapid detection of oil palm trees under complex conditions such as overlapping canopy, different tree structures and varying canopy sizes. It can not only provide accurate basic data for oil palm tree counting, yield and carbon storage estimation, but also provide technical guidance for other forest types, such as eucalyptus and pine, and single tree detection and segmentation. |
|---|---|
| ISSN: | 1574-9541 |