CoSimR: Similarity Cues-Aware Consanguinity Relationship Mining Framework for IoT-Based Bird Monitoring Systems

Internet of Things (IoT)-based bird monitoring system faces challenges from occlusion, arbitrary postures, and similar species. To address these, we exploit two key properties of bird images: self-correlation within individual birds and appearance similarity across species, revealing self-correlation and consanguinity relationships. We propose a similarity cues-aware consanguinity relationship mining (CoSimR) framework, which leverages these relationships for robust classification. CoSimR comprises two modules: Consanguinity Relationship Mining (CRM) and Cross-species Avian Prediction (CAP). CRM can capture skeletal structures by generating the correlation tokens, while consanguinity tokens encode p information across five taxonomic levels (class, order, family, genus, species). CAP leverages a consanguinity-driven multi-loss function, including homogeneity loss for intra-species consistency and affinity loss for cross-species similarity, to guide discriminative feature learning. Experiments conducted on two fine-grained bird image classification datasets demonstrate that the CoSimR model achieves better performance compared with state-of-the-art methods. It highlights the effectiveness of integrating biological hierarchy with visual features, paving the way for similarity cues-aware approaches in fine-grained classification tasks.