Object detection algorithm for eggs of Pomacea canaliculata in a paddy field environment

As an invasive species in China, Pomacea canaliculata severely impacts crop quality and yield, necessitating effective monitoring for food security. To address the challenges in detecting its eggs in paddy fields-including feature contamination, stem and leaf occlusion, and dense targets-we propose an enhanced YOLOv8n-based algorithm. The method introduces omni-dimensional dynamic convolution (ODConv) in the backbone network to improve target feature extraction, constructs a Slim-neck structure to optimize feature processing efficiency, and designs a receptive-field attention head (RFAHead) for detection refinement. Experimental results demonstrate that the improved model achieves 3.3% and 4.2% higher mAP@0.5 and mAP@0.5:0.95 than the original YOLOv8. It outperforms Faster R-CNN, YOLOv3-tiny, YOLOv5, YOLOv6, YOLOv7-tiny, YOLOv9-t, YOLOv10n, and YOLOv11n by 18.2%, 12.4%, 5.2%, 10.8%, 11.6% 5.0%, 3.8%, and 3.4% in mAP@0.5 and 20.6%, 17.5%, 8.1%, 15.6%, 16.1%, 7.0%, 7.7%, and 6.5% in mAP@0.5:0.95, respectively. Visual analysis confirms enhanced recognition of small and occluded targets through improved feature learning. This model enables accurate and rapid detection of Pomacea eggs in rice fields, offering technical support for invasive species control.