UDA-FlyRecog: Unsupervised domain adaptation for drosophila cross-domain recognition model

Abstract

Drosophila is an important quarantine pest, and its identification plays a crucial role in agricultural development. To address the issue of significant style variations in Drosophila image samples collected from different environments, which results in poor performance when applying models trained on the source domain to the target domain, this paper proposes a domain-adaptive model for cross-domain Drosophila recognition (UDA-FlyRecog). This model combines global feature alignment and class-aware feature alignment methods, and adopts a cyclic iterative training approach to mitigate the effects of domain shift. Experimental results demonstrate that our method achieves the highest accuracy improvement of up to 16.7% compared to other methods on Drosophila datasets collected from both laboratory and natural environments, indicating promising practical applications.