Marine Organism Detection Algorithm Based on MSPSO-YOLOv8

To address critical challenges in underwater organism detection, including poor small-target recognition performance and limited detection precision, we present an enhanced MSPSO-YOLOv8 architecture. This algorithm achieves performance improvement through three technological innovations: Firstly, the SPDConv spatially depth-separable convolution module is adopted to enhance the detail retention ability during the feature downsampling process; Secondly, integrating the large kernel Separable Attention Mechanism (LSKA) enhances the multi-scale feature focusing ability of the model; Finally, an innovative multi-scale pyramid optimization structure (MSPSO) was designed to improve the positioning accuracy of micro-targets. Test results on the URPC2020 benchmark dataset show that the method achieves 82.6% accuracy of [email protected], a 3.1% improvement over the original model, and demonstrates significant advantage particularly in the recognition of sub-pixel microorganisms.