In the domain of underwater object detection, real-time performance and the utilization of lightweight models are critical requirements. In order to address these specific demands, we propose PGNet, an innovative and lightweight backbone architecture that efficiently reduces the computational requirements and inference time of the model, while ensuring a high level of accuracy. We conducted experiments on two different underwater image datasets. On the brackish-underwater dataset, our design surpasses existing lightweight backbones by achieving a parameter count of 2.01 million and consuming merely 4.5 GFLOPs. When evaluated on the validation set, our model attains an impressive Mean Average Precision (Map) score of 0.968, accompanied by a remarkable frame rate of 318 frames per second (FPS). These results unequivocally demonstrate the exceptional performance and efficiency of our proposed approach in the context of underwater object detection. PGNet holds significant potential for practical applications in areas such as marine science, underwater engineering, and underwater rescue. The codes and trained models will be available soon.