Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study focuses on lightweight classification of raptor species for real-time deployment on edge devices, specifically targeting collision mitigation for wind turbines. By employing knowledge distillation from a DINOv2-L teacher model to three lightweight student models and significantly expanding the dataset to 12,519 images, the researchers achieved a macro recall of 0.935 with a reduced parameter count. Notably, the improved dataset led to a 38.6 percentage point increase in recall for the White-tailed Eagle, while drastically reducing misclassification rates with a TensorRT deployment achieving high throughput on edge hardware.
Expanding the dataset for raptor classification led to a remarkable 38.6 percentage point boost in recall for the White-tailed Eagle, showcasing the power of data in model performance.
We investigate lightweight raptor-species classification for real-time edge deployment in wind-turbine collision mitigation. Using DINOv2-L (304M parameters) as a teacher, we distilled three lightweight students (MobileNetV4, ViT-Small, and EfficientNet-B0). To reduce confusion between closely related species, we expanded the dataset to 12,519 images, including an increase in Steller's Sea Eagle images from 463 to 2,050 via video-frame extraction. Under a group split that separates samples at the video- and source-image level to mitigate source leakage at that granularity, the three-student ensemble achieved a macro recall of 0.935 +/- 0.004 over five distillation seeds (0.955 on a conventional image-level split, retaining 97.5% of the teacher's macro recall) with roughly one-eighth as many parameters. On a subset of 1,258 images disjoint from the former training images, White-tailed Eagle recall improved by up to 38.6 percentage points, while the rate at which it was misclassified as the Steller's Sea Eagle decreased from 61% to 15% of errors. TensorRT FP16 deployment of EfficientNet-B0 on an NVIDIA Jetson Orin Nano achieved 3.19 ms/image including host-device transfer (313 images/s), with 99.95% argmax agreement with FP32. In five-seed controlled comparisons, neither distillation (versus CE-only) nor the change from a DINOv2-L to a DINOv3-L teacher yielded a clear ensemble-level improvement; the primary gains stem from the dataset expansion and teacher re-fine-tuning.