Laser-based machine vision and volume segmentation algorithm for equal weight portioning of poultry and fish

Automation in meat portioning plays a critical role in improving efficiency, accuracy, and consistency in food processing. Poultry and fish industries face challenges in achieving equal-weight portions due to irregular shapes, variable thickness, and inconsistent density, which affect segmentation accuracy. Addressing these issues is essential for reducing material waste, ensuring compliance with industry tolerances, and improving yield consistency. This study aimed to design, develop, and evaluate a laser-based machine vision system and volume segmentation algorithm capable of producing equal-weight portions of poultry and fish under realistic processing conditions. The proposed system employed a laser line profile sensor to capture three-dimensional point cloud data of meat surfaces. Data were processed using a Delaunay triangulation-based approach to compute object volume, assuming uniform density. An iterative volume segmentation algorithm determined optimal vertical and horizontal cutting lines to match target volumes. The system was tested on both modeling clay and real meat samples, including chicken breast and fish fillets. Statistical analyses included mean absolute error (MAE) to assess weight deviation, one-way analysis of variance (ANOVA) to evaluate differences between cutting patterns, and Gage Repeatability & Reproducibility (Gage R&R) to assess measurement system reliability. For real meat tests, chicken breast portioning yielded %MAE between 4.53% and 5.96%, while fish fillet portioning achieved %MAE between 3.63% and 5.78%. Gage R&R analysis showed total measurement system variation below 10%, confirming high repeatability and reproducibility. Fish fillets exhibited lower variability due to flatter morphology and more complete point cloud acquisition. Modeling clay trials demonstrated consistent accuracy across different cutting configurations, with ANOVA revealing statistically significant differences between certain patterns. The developed laser-based machine vision system with volume segmentation reliably produced equal-weight portions for irregularly shaped raw meat products. The method-maintained accuracy within industry-accepted tolerances for both chicken and fish, while exhibiting robust performance under variations in sample placement and geometry. Future work should focus on increasing point cloud density for complex surfaces, integrating deep learning for adaptive cutting line adjustment based on variations in thickness and density, and extending the approach to other food products such as dairy and bakery items to broaden its industrial applicability.