Abstract: Population census of sea cucumbers serves as a fundamental operational metric in contemporary aquaculture, providing essential data for production management, feed administration, and commercial transactions. Despite its critical importance, the industry continues to rely predominantly on manual sampling counting methods, which present multiple operational challenges. These conventional approaches suffer from three principal limitations: 1) exorbitant labor requirements leading to unsustainable operational costs; 2) suboptimal counting speeds resulting in processing bottlenecks during high-volume operations; 3) inherent human error factors causing significant measurement inaccuracies. The continuous expansion of sea cucumber farming has exacerbated these issues, with current production scales rendering manual methods economically and logistically impractical. This technological gap creates an urgent industry demand for automated, intelligent counting solutions capable of delivering accurate, real-time population data. To address these critical challenges, we present an innovative intelligent counting framework that leverages advanced computer vision technologies deployed on unmanned surface vehicles (USVs). Our integrated solution framework comprises three synergistic technical modules: object detection, target tracking, and automated counting. For the detection module, we developed YOLOv8s-BB, an optimized detection model that significantly improves underwater object recognition through two key architectural innovations. First, we designed the BiFormer Convolutional Attention Module (BCAM) by strategically combining a bi-Level routing attention mechanism with spatial attention mechanism. This hybrid attention module was then integrated at the terminal layer of the YOLOv8s Backbone network. Second, we inserted dedicated BiFormer modules after each C2f block in the Neck architecture. These modifications collectively enhance feature extraction and fusion capabilities, particularly for challenging underwater environments with variable lighting conditions and turbidity. In the tracking module, we introduce TriSORT, an advanced tracking algorithm featuring triple-matching that effectively addresses the target association limitations of conventional ByteTrack in aquatic settings. Specifically, the TriSORT algorithm implements a three-stage cascade matching process between detection boxes and predicted boxes, effectively leveraging high-confidence detection results to enhance tracking accuracy and robustness. The counting module implemented and rigorously compared two counting methodologies: traditional line-crossing method and our innovative inactive trajectory removal method. The latter method demonstrates superior performance by dynamically filtering transient detections and maintaining only validated trajectories, thereby eliminating common counting artifacts prevalent in aquatic environments. Comprehensive experimental evaluations demonstrate the superiority of our framwork: (1) The YOLOv8s-BB detector achieves state-of-the-art performance with 88.9% mean average precision (mAP), 77.8% recall, and 84.2% F1 score, outperforming existing YOLO model variants: YOLOv8s, YOLOv7-tiny, YOLOv9s, and YOLOv11s models; (2) The TriSORT tracking algorithm exhibits enhanced tracking robustness, attaining 74.00% multiple object tracking accuracy (MOTA) and 85.03% IDF1 score, representing improvements of 6.55 and 5.54 percentage points respectively over ByteTrack; (3)The inactive trajectory removal counting method achieved superior performance with an average counting accuracy of 95.46% and a mean absolute error of 1.90, demonstrating superior performance over conventional line-crossing counting methods in all evaluated metrics. This study makes an important contribution to aquaculture technology by establishing a fully optimized detection-tracking-counting pipeline that enables precise, efficient, and automated sea cucumber population assessment. The developed system provides reliable, data-driven support for critical aquaculture operations including biomass estimation, feeding optimization, and harvest planning, with potential applications extending to other marine species monitoring.