摘要： 鱿鱼是中国的重要水产品，但其加工行业仍依赖劳动密集型操作，存在生产效率低，劳动强度大等问题。本研究提出了一种基于图像识别的柔体建模方法，利用鱿鱼胴体的二维图像生成待加工时悬垂充盈的胴体三维模型，为鱿鱼加工实现自动化提供模型基础。具体方法是先用OpenCV进行图像识别，提取胴体的边缘轮廓及几何信息，再用MATLAB对其进行计算与调整，推导内腔轮廓点与旋转状态胴体轮廓点，并将其绕中心轴旋转生成网格模型。设置两组数据集进行验证，包括同一胴体的不同姿势与5条大小不同的鱿鱼胴体，结果表明胴体模型外轮廓线长度与胴体实际尺寸误差最大为0.66%，3种姿势下生成的模型之间几何尺寸误差都在1%内，证明了该方法的准确性，能适应不同姿势的平展胴体二维图像，生成其待加工时的三维模型。

关键词: 鱿鱼胴体, 柔体平展, 图像识别, 三维重建, 姿势对比

Abstract: Squid is a significant aquatic product in China, yet its processing industry remains labor-intensive, leading to low efficiency and high labor intensity. This study introduces an image recognition-based soft-body modeling method to generate a 3D model of a suspended and inflated squid mantle from 2D images, providing a foundation for automated squid processing. The method uses OpenCV for image recognition to extract edge contours and geometric features of the squid mantle. MATLAB then refines these contours, deduces inner cavity contour points, and determines the mantle's rotational state. The contours are rotated around the central axis to construct a mesh model. This process accurately captures the shape of the squid mantle during processing, accounting for its deformable and flexible nature. Validation was performed using two datasets: different postures of the same squid body and five squid bodies of varying sizes. Results show that the maximum error in the model’s outer contour length compared to the actual body size is 0.66%, and geometric size errors across three postures are within 1%. This confirms the method's accuracy in generating 3D models from 2D images of squid bodies in various poses. The method ensures precise contour representation and is suitable for automated path planning in tasks like cleaning and cutting. Its non-contact nature makes it ideal for food processing, avoiding radiation issues associated with CT scanning. This study lays a foundation for automated squid processing, supporting tasks such as cleaning, cutting, and grading.

Key words: squid carcass, soft-body flattening, image recognition, 3D reconstruction, pose comparison