Abstract: In high-density aquaculture systems, the overlap and occlusion among Litopenaeus vannamei individuals, coupled with interference from the background such as residual feed and debris, suggest that not all visible individuals are suitable for measurement and body mass estimation. Meanwhile, the species' semi-transparent body surface and slender contours further increase the difficulty of directly measuring individuals and estimating body mass from full-view images. To obtain measurement-qualified individuals from such complex environments and accurately estimate the average body mass, we adopted a two-stage strategy of "screening first, segmentation later". During the dataset construction phase, individuals with clear structures and complete postures were annotated as "measurable individuals", and the YOLOv11 model was employed to detect and filter measurable individuals from full-view images. Subsequently, we constructed an edge-aware fused segmentation network to enhance the recognition of semi-transparent structures and fine contours, enabling high-precision individual segmentation and geometric parameter extraction. The shrimp body length and width were calculated using a minimum bounding rectangle strategy, and the weight was estimated via a power-function regression model. In practical experiments, the proposed segmentation model achieved a Dice score of 0.898 6, with a mean absolute error of 2.46 mm in length measurement and 0.51 g in body mass prediction. The results demonstrate that the proposed method can achieve precise non-contact measurement of farmed L. vannamei in clear-water, factory-based recirculating aquaculture systems, making it a promising tool for intelligent aquaculture management.