Abstract: In modern land-based intensive aquaculture systems, the real-time and accurate estimation of live fish weight plays a critical role in precision feeding, growth monitoring, and overall farm management. However, environmental constraints such as high stocking density, limited camera viewpoints, water turbidity, and fish movement often result in partial occlusion, where only the fish head is visible, complicating conventional visual measurement approaches. To address this challenge, this study investigates two methodologies for estimating the weight of largemouth bass (Micropterus salmoides) under head-visible and body-occluded conditions. The first approach directly predicts fish weight based on morphological features extracted from the visible head region, leveraging linear and nonlinear regression models trained on head-length and width measurements. While this method provides a straightforward and computationally efficient estimation, it is inherently limited by the absence of complete body information, which introduces significant errors under occlusion scenarios. To overcome these limitations, the second approach introduces a novel two-stage generative completion framework. In the first stage, the partially visible head image is input into a generative model designed to synthesize the complete fish body, reconstructing previously unobserved morphological structures. The second stage utilizes the generated full-body image to extract comprehensive morphological features, including body length, height, and lateral area, which are then used to estimate fish weight through multivariate regression models. Experimental evaluations conducted on two largemouth bass datasets demonstrate that the proposed two-stage method substantially improves estimation accuracy. Specifically, the mean absolute error (MAE) decreased from an average of 83 grams using the direct head-based estimation to 63 grams, and the mean absolute percentage error (MAPE) was reduced from approximately 30% to 22%. The results indicate that the generative completion process effectively compensates for missing morphological information caused by occlusion, providing not only quantitative weight estimates but also a visual, interpretable reference for the reconstructed fish body. Compared to the direct estimation method, this framework offers a robust and intuitive solution for biomass monitoring in intensive aquaculture systems, enhancing the reliability of automated growth assessment, enabling informed feeding decisions, and supporting the integration of computer vision techniques into smart aquaculture management. Furthermore, the generative stage facilitates visualization under extreme occlusion conditions, allowing farm operators and monitoring systems to assess fish morphology and size even when direct observation is partially obstructed. The findings underscore the potential of combining generative image completion with stereo vision-based measurement and regression modeling to address practical challenges in aquaculture environments, offering a promising direction for real-time, non-invasive fish biomass estimation and intelligent farm operation.