This paper addresses the challenge of efficient small-object detection in underwater images, which is crucial for marine ecosystem research and species conservation. The authors propose an enhanced version of the You Only Look Once Version 8 (YOLOv8) model, specifically optimized for underwater scenarios. The key contributions include: 1. **Model Lightweighting**: Replacing the Darknet-53 backbone with FasterNet-T0 reduces model parameters by 22.52%, FLOPS by 23.59%, and model size by 22.73%, achieving a lightweight design. 2. **Enhanced Small-Object Detection**: Adding a Prediction Head for Small Objects and adjusting the number of channels in high-resolution and low-resolution feature maps improves small-object detection accuracy by 1.2%. 3. **Improved Detection Accuracy**: Utilizing Deformable ConvNets and Coordinate Attention in the neck part enhances the detection of irregularly shaped and densely occluded small targets. The proposed method achieves 52.12% AP on the UTDAC2020 underwater dataset, surpassing the performance of the larger YOLOv8l model (51.69% AP) with significantly fewer parameters and computational resources. When increasing the input image resolution to 1280 × 1280 pixels, the model achieves 53.18% AP, making it the state-of-the-art (SOTA) model for the UTDAC2020 dataset. Additionally, the method achieves 84.4% mAP on the Pascal VOC dataset, demonstrating its effectiveness and generalization to common datasets. The paper also includes a detailed experimental setup, comparisons with other state-of-the-art methods, and an ablation study to validate the effectiveness of each component. The results show that the proposed method maintains high accuracy while significantly reducing the model's complexity, making it suitable for resource-constrained underwater applications.This paper addresses the challenge of efficient small-object detection in underwater images, which is crucial for marine ecosystem research and species conservation. The authors propose an enhanced version of the You Only Look Once Version 8 (YOLOv8) model, specifically optimized for underwater scenarios. The key contributions include: 1. **Model Lightweighting**: Replacing the Darknet-53 backbone with FasterNet-T0 reduces model parameters by 22.52%, FLOPS by 23.59%, and model size by 22.73%, achieving a lightweight design. 2. **Enhanced Small-Object Detection**: Adding a Prediction Head for Small Objects and adjusting the number of channels in high-resolution and low-resolution feature maps improves small-object detection accuracy by 1.2%. 3. **Improved Detection Accuracy**: Utilizing Deformable ConvNets and Coordinate Attention in the neck part enhances the detection of irregularly shaped and densely occluded small targets. The proposed method achieves 52.12% AP on the UTDAC2020 underwater dataset, surpassing the performance of the larger YOLOv8l model (51.69% AP) with significantly fewer parameters and computational resources. When increasing the input image resolution to 1280 × 1280 pixels, the model achieves 53.18% AP, making it the state-of-the-art (SOTA) model for the UTDAC2020 dataset. Additionally, the method achieves 84.4% mAP on the Pascal VOC dataset, demonstrating its effectiveness and generalization to common datasets. The paper also includes a detailed experimental setup, comparisons with other state-of-the-art methods, and an ablation study to validate the effectiveness of each component. The results show that the proposed method maintains high accuracy while significantly reducing the model's complexity, making it suitable for resource-constrained underwater applications.