A knowledge-spatiotemporal integrated acoustic recognition framework for ecological monitoring

Abstract

Bioacoustic monitoring is crucial for biodiversity assessment and ecological conservation. Traditional manual methods are inefficient, while existing deep learning-based models suffer from unexplainable recognition, high false alarms, insufficient use of ecological prior knowledge, and poor integration with monitoring operations. This paper proposes the Knowledge-Spatiotemporal Integrated Bird Vocalization Recognition Framework (KSIBird), which integrates knowledge graph reasoning with spatiotemporal constraint validation. It constructs an avian ecological knowledge graph leveraging species-habitat-distribution relationships for reasoning-based recognition correction, and introduces a spatiotemporal constraint model incorporating migratory routes and seasonal windows for consistency verification, forming a triple-tiered mechanism: acoustic feature recognition, knowledge graph reasoning, and spatiotemporal constraint validation. Experiments on 200 field-recorded samples show accuracy improves from 83.5% to 91.0% and false alarm rate drops from 12.0% to 6.5%. The proposed method offers an explainable, verifiable, and evolvable computational framework for bioacoustic monitoring.