基于多模态融合的东星斑低温异常行为识别OA

The behavioral responses of the eastern red snapper(Plectropomus leopardus)to low-temperature stress are of significant importance for the precise regulation of aquaculture environments.This study proposes a multimodal-structured classification model for recognizing the behavior of eastern red snapper under low-temperature stress,based on the spatiotemporal distribution differences observed under such conditions.The model integrates three modalities,including RGB images,binarized images,and optical flow images,to improve behavioral classification accuracy.It employs three independent ResNet-50 feature extraction modules and introduces an attention-based fusion strategy to dynamically adjust the contributions of different modal features.Experimental results demonstrate that the three-modal fusion model achieves an accuracy of 96.5%,a precision of 96.8%,a recall of 97.2%,and an F1 score of 97.4%,outperforming other comparative models.These results validate the complementarity and effectiveness of multimodal features in behavior recognition.This study provides a quantitative analysis of behavioral changes in eastern red snapper under low-temperature stress and establishes a methodological foundation for implementing behavior monitoring systems in aquaculture environments.