While Unmanned Aerial Vehicles (UAVs) offer promising capabilities for aerial monitoring, there remains a critical shortage of publicly available, UAV-acquired datasets specifically designed for machine learning-based intrusion detection in agricultural settings. Existing datasets either lack aerial perspectives, omit essential “empty space” categories for negative sampling, or are not optimized for ternary classification tasks required for robust farm security systems. To bridge this gap, this paper introduces a novel UAV-acquired dataset specifically designed for farm intrusion detection. Collected through systematic aerial surveillance supplemented with ground-level captures, the dataset comprises 2,067 images across three essential categories: animals (907 images), people (588 images), and empty spaces (572 images). The inclusion of empty spaces enables models to distinguish between normal and intrusion scenarios, while UAV-captured aerial views provide comprehensive coverage ideal for large-scale farm monitoring. We detail the data collection methodology using consumer-grade UAVs, preprocessing techniques, and dataset characteristics, with visualizations confirming its balanced distribution and suitability for real-world applications. To demonstrate practical utility, We provide an initial baseline model which achieves a 98.4% accuracy and full documentation to facilitate reuse and extension by the community. The dataset and related codes are publicly available at https://github.com/mugishastanley/Intrusion_detection to support further research in UAV-based agricultural security and represents a significant contribution toward automated intrusion detection systems in the era of Agriculture 4.0.
