<p>Accurate measurement of fish body length, height, and thickness is crucial for aquaculture management; however, existing methods often suffer from limited accuracy and are incapable of capturing thickness information. In order to enhance the precision of measurement locations, this study proposes a binocular stereo vision-based approach for fish body size estimation. The framework integrates pixel-level keypoint localization with depth perception and utilizes a Ghost module branch to eliminate redundant convolutional features, while a channel shuffle mechanism is proposed to enhance keypoint correlations and to robustly handle limited disturbances, such as minor illumination variations. Furthermore, binocular stereo matching is utilized to reconstruct three-dimensional structural information, enabling the conversion of two-dimensional pixel coordinates into three-dimensional spatial coordinates for precise measurement of body length and height and leveraging depth data to overcome the limitation of thickness measurement. A large yellow croaker dataset comprising individuals of varying sizes was constructed to evaluate the method. Experimental results demonstrate that the proposed approach achieves a detection accuracy of 99.5% (mAP50–95) on the dataset, with body size estimation errors maintained within ± 5%. These findings highlight the effectiveness of the method for high-precision measurement and its potential to provide reliable data support for scientific breeding and feeding strategy optimization.</p>