LAT-BirdDrone

Currently, a notable limitation in existing datasets is their predominant focus on object classification based on appearance. Specifically, datasets tailored for trajectory classification of low-altitude flying micro-targets (with a size of 11pt × 11pt or smaller) are extremely scarce. Moreover, there is an urgent demand for a dedicated dataset that can provide comprehensive data support to verify the impact of the CNN module on trajectory classification models.To address these critical gaps, this research leveraged the Three-band refrigeration photoelectric turntable Z50IV-CTVC690110-21100—an advanced long-distance optoelectronic detection system independently developed by Hepu Vision—for data collection. The data acquisition process covered multiple videos, including both visible light and infrared footage of birds and drones, ensuring the diversity of data sources.In the data processing phase, the YOLOv11n algorithm was applied to conduct object detection and trajectory extraction. To further enhance the accuracy and reliability of target trajectory extraction, the ByteTrack algorithm was additionally employed, forming a robust data processing pipeline.Through the aforementioned efforts, the LAT-BirdDrone dataset—dedicated to the trajectory classification of low-altitude flying micro-targets—was successfully established. This dataset comprises 33,262 image samples and 665 pieces of detailed trajectory data. It comprehensively covers the dynamic motion patterns of low-altitude flying micro-targets (11pt × 11pt or smaller) under various environmental conditions, including different lighting intensities, weather scenarios, and complex background settings.In terms of reuse potential, the LAT-BirdDrone dataset first fills the scarcity of datasets specialized in trajectory classification of low-altitude flying micro-targets (11pt × 11pt or smaller). Secondly, it can provide essential data support for verifying the impact of the CNN module on trajectory classification models. Thirdly, it serves as a valuable resource for evaluating the trajectory classification performance of basic models such as CNN, Transformer, iTransformer, and LSTM. Furthermore, it supports research on multi-modal hybrid architectures (e.g., Transformer+LSTM, CNN+BiLSTM) that integrate local spatial features with global temporal dependencies. Additionally, the dataset offers necessary resources for studies on micro-targets in low-altitude security applications and provides references for optimizing the application of optoelectronic sensors in low-altitude micro-target monitoring scenarios.