Underwater Treasure Object Detection Algorithm Based on Improved YOLOv8

For the target detection algorithm of undersea treasure graphics (including underwater sea cucumber, sea urchin and scallop). In the current field of target detection, some problems such as high model complexity, low detection efficiency and weak generalization ability, so a new object detection algorithm of underwater treasure images based on improved YOLOv8 is proposed. On the one hand, in view of the fuzzy and small size of the underwater treasure image target, the attention mechanism of YOLOv8 can be lightweight to cut down the calculation amount in the model and increase the efficiency. SDP-Conv was lead in backbone and neck network to enhance feature extraction capability. Deformable convolution is introduced into the backbone network to boost the detection precision. The experimental results on Zhanjiang underwater robot competition data set show that the highest accuracy of YOLOv8 algorithm model is grew to 98.9% by the optimization algorithm. Compared to other basic algorithms, the proposed algorithm greatly reduces the model volume, parameter number and calculation amount while ensuring high detection accuracy. From the results we can see that the proposed algorithm achieves the balance between model complexity and detection accuracy.