ResAF-Net: Anchor-Free AI for Tree Detection in Palestine
A new satellite-based tree detection framework, ResAF-Net, has been developed for large-scale agricultural monitoring in Palestine, where fragmented landscapes and restrictions on aerial monitoring hinder data collection. The architecture combines a ResNet-50 encoder, Atrous Spatial Pyramid Pooling, feature fusion, multi-head self-attention, and an anchor-free FCOS detection head. Trained on the MillionTrees benchmark, it achieved 82% Recall, 63.03% mAP@0.50, and 35.47% mAP@0.50:0.95 on validation, showing strong sensitivity to tree presence.
Key facts
- ResAF-Net is a satellite-based tree detection framework for agricultural monitoring in Palestine.
- The architecture uses ResNet-50, ASPP, feature fusion, multi-head self-attention, and FCOS head.
- Trained on MillionTrees benchmark.
- Achieved 82% Recall, 63.03% mAP@0.50, 35.47% mAP@0.50:0.95 on validation.
- Designed for resource-constrained settings with limited field access and aerial monitoring restrictions.
- Aims to improve tree localization in dense and heterogeneous scenes.
- Published on arXiv with ID 2604.23653.
- Addresses food security, land-use planning, and economic resilience in Palestine.
Entities
Institutions
- arXiv
Locations
- Palestine