Underwater object detection is challenging due to color distortion, low contrast, turbidity, noise, complex backgrounds, and partial occlusion, which can reduce the effectiveness of conventional object detection models. This study proposed an Enhanced AquaMamba-YOLO framework that integrated CNN-based local feature extraction, Mamba-based global contextual modeling, multi-scale feature fusion, and YOLO-based detection for improved underwater object detection. The input images were processed using color correction, Contrast Limited Adaptive Histogram Equalization (CLAHE), normalization, and data augmentation to improve visual quality and robustness. The CNN backbone extracted hierarchical local features, while the Mamba module modeled long-range contextual dependencies. These complementary representations were subsequently integrated through multi-scale feature fusion and provided to the YOLO v8 detection head for object classification and bounding-box localization. The framework was evaluated on the Aquarium Data COTS dataset using precision, recall, F1-score, mAP@0.5, mAP@0.5:0.95, and Frames Per Second (FPS). The proposed model achieved 0.954 precision, 0.935 recall, 0.963 mAP@0.5, 0.735 mAP@0.5:0.95, and 0.944 F1-score, while maintaining an inference speed of 94 FPS. The ablation results demonstrated progressive improvements with the addition of image enhancement, CNN feature extraction, Mamba, and multi-scale fusion modules. These results indicated that the proposed framework effectively improved underwater object detection while maintaining practical computational efficiency.

Underwater Object Detection, Convolutional Neural Network, Mamba Network, YOLO v8, Multi-Scale Feature Fusion, Image Enhancement