Underwater trash detection is a key technology for underwater robots to handle underwater trash. However, the complexity and variability of the underwater environment and poor lighting conditions, as well as the loss of detail information and poor performance of low-resolution images due to step-length convolution in traditional CNN models, limit the accuracy and speed of existing models. In order to solve these problems, this paper proposes a novel underwater rubbish detection algorithm SPDC-YOLOv8. The algorithm employs an adaptive spatial decomposition-based CNN module SPD-Conv in the backbone network of YOLOv8, replacing the step-length convolution, thus improving the accuracy of the model for low-resolution images and small object detection. Meanwhile, the CARAFE operator is used in the up-sampling process of the model, which enhances the semantic information and feature representation of underwater rubbish, and thus improves the robustness of Object Detection. The experimental results show that the method proposed in this paper obtains 98.6% and 91.2% accuracy on the trash_ICRA19 dataset and TrashCan dataset, respectively, and improves 0.3% and 0.8% compared with the original YOLOv8 model, and the computation time is 2.6 ms in both cases. The improved YOLOv8 algorithm proposed in this paper is more adapted to the underwater complex environment underwater environment.