Underwater object detection is the complex task of identifying and locating objects within aquatic environments using underwater cameras. This study evaluates the effectiveness of YOLOv8 and YOLOv9 object detection algorithms in underwater environments typical of aquaculture settings. Our research aims to enhance the detection accuracy, speed, and reliability of the YOLO (You Look Only Once) models. Contrast Limited Adaptive Histogram Equalization (CLAHE) and Dark Channel Prior (DCP) were evaluated on the dataset to enhance the images. Performance was assessed using mean Average Precision (mAP), precision, and recall. Results show that YOLOv9, with its anchor-free design and optimized processing, outperforms YOLOv8, in handling overlapping objects and reducing false positives. Compared to YOLOv8, YOLOv9 demonstrated impressive mAP values of 0.918 for mAP@50 and 0.538 for mAP@95, with a precision of 0.899 and a recall of 0.852. We analyze various tuning parameters to identify the optimal settings. An ablation study is performed to assess model performance with and without preprocessing steps. These findings contribute insights into the application of advanced deep learning in aquaculture, highlighting significant improvements in underwater object detection technology.