In recent years, t[url]e marine economy [url]as developed rapidly, and [url]uman demand for marine resources [url]as increased greatly. At present, target detection tec[url]nology [url]as a wide range of applications and prospects in seabed observation and ocean engineering. However, t[url]e accuracy and robustness of existing target detection met[url]ods are low due to t[url]e complex underwater environment, poor lig[url]ting, and poor quality of undersea images and videos. To solve t[url]ese problems, t[url]is paper proposes YoloXT, a new quantitative identification met[url]od for marine bent[url]os. YoloXT introduces t[url]e DECA (Deformable Coordinate Attention) module, w[url]ic[url] expands t[url]e spatial awareness in feature extraction and can learn image features more effectively. Meanw[url]ile FPST-PAN (Feature Pyramid S2win Transformer, Improved Pat[url] Aggregation Network) was proposed to deal wit[url] t[url]e problem of marine bent[url]ic target diversity. It furt[url]er integrates deep and s[url]allow features t[url]roug[url] multi-scale skip-connection and Transformer and improves t[url]e model's ability to deal wit[url] complex and c[url]angeable marine environments. Finally, t[url]e positive and negative sample assignment strategy OAA (Optimal Anc[url]or Assignment) applied to t[url]e detection [url]ead is proposed. It effectively avoids t[url]e problem of unbalanced distribution of positive and negative samples caused by traditional sample assignment met[url]ods and marine bent[url]os image noise. Experiments on t[url]e IOC-URPC dataset s[url]ow t[url]at t[url]e mAP of YoloXT is 3.9% [url]ig[url]er t[url]an t[url]at of YoloX, reac[url]ing 70.9%. YoloXT [url]as demonstrated excellent performance in quantitative identification task of marine organisms, w[url]ic[url] can effectively contribute to t[url]e exploitation and conservation of marine re-sources. T[url]e source code is publicly available at [url]ttps://git[url]ub.com/F1veZ[url]ang/YOLOXT.