Sex Detection in Poecilia Reticulata Based on Dual-Modal

Poecilia reticulata is often used as a research object in the laboratory because of its short breeding cycle and impressive reproductive ability, and sex detection can be used to carry out reasonable sex rationing and breeding density control to improve feed utilization and breeding yield, However, because it is more difficult to distinguish the sex of Poecilia reticulata artificially, we proposed a bimodal based sex detection algorithm for Poecilia reticulata. In this algorithm a bimodal image alignment method is used, where the captured RGB video is recorded with the event camera and then the bimodal images are aligned with perspective transformation. In our method, we used the pixel level fusion of yolov11, introduced CBAM attention, Poolformer in Metaformer improved C3K2 module, map@50 reached 94.9 %, Recall reached 90 %, which is 5.7 % improvement compared to YOLOv5n Recall, and MAP50 improved by 2.4%, compared to YOLOV8n Recall improved by $\mathbf{2. 3 \%}$, MAP50 improved by 1.2 %, compared to YOLOv10n Recall improved by 3.4 %, MAP50 improved by 1.5 %, YOLOV11 Recall improved by 1.8 %, MAP50 improved by 0.6 %, which can detect the sex of Poecilia reticulata accurately.