Abstract: In order to solve the problems of low efficiency and significant stress damage in traditional manual biomass monitoring during the cultivation of Pacific white shrimp (Litopenaeus vannamei), we propose a non-contact body-length and mass estimation method that integrates an improved YOLOv8 detector with an oriented bounding box (OBB) framework, so as to improve the efficiency and accuracy of intelligent feeding management. First, we constructed a Focal-GIoU loss function to bolster detection performance under heavy occlusion. Then, we adopted a GhostNetV2 backbone enhanced with a spatial-channel decoupling attention mechanism and integrate dynamically reconfigurable RepBlock modules to strengthen multi-scale morphological adaptation. Finally, we introduced a Principal Component Analysis-Oriented Bounding Box (PCA-OBB) algorithm in which the shrimp's principal axis was extracted via eigen decomposition of its covariance matrix and a Hough-circle detection scheme calibrated the pixel-to-physical dimension conversion coefficient. A regression model which correlated body length to mass was established. Experiments on the 120-day-old L. vannamei samples demonstrated an average relative error of 1.26% in length measurement, maximum absolute error of 0.503 cm, and a 5.3% average relative error in mass prediction, both outperforming conventional manual methods. This method achieves real-time monitoring and estimation of non-contact biomass parameters of L. vannamei, providing effective technical support for precise feeding.