包装与食品机械2025，Vol.43Issue(1)：17-25,9.DOI:10.3969/j.issn.1005-1295.2025.01.003
基于改进Mask R-CNN的对虾部位分割方法
Shrimp body part segmentation method based on improved Mask R-CNN
摘要
Abstract
To address the challenge of precise part segmentation in shrimp processing,a deep learning-based method was proposed for semantic segmentation of shrimp body parts.The Simple Attention Module(SimAM)was integrated into the residual blocks of ResNet,the feature extraction network within the Mask R-CNN model.Additionally,the Sobel operator was employed for edge extraction,and an edge loss function was incorporated to enhance edge segmentation accuracy.Results demonstrated that the proposed model achieved a mean Intersection over Union(mIoU)of 94.14%,mean Pixel Accuracy(mPA)of 97.06%,and part-specific mIoU values of 84.12%for the head,83.79%for the thorax,and 94.31%for the tail.Comparative experiments with UNet,PSPNet,SegNet,and SegFormer under the same dataset and experimental conditions confirmed the superior segmentation performance of the proposed method.This study provides a novel approach for shrimp processing.关键词
对虾/部位分割/Mask R-CNN模型/边缘监督/注意力机制Key words
shrimp/part segmentation/Mask R-CNN model/edge supervision/attention mechanism分类
轻工业引用本文复制引用
蔡礼扬,宁萌,杨九洲,陈义亮,马泓睿,王雨芊..基于改进Mask R-CNN的对虾部位分割方法[J].包装与食品机械,2025,43(1):17-25,9.基金项目
国家重点研发计划项目(2022YFD2100304) （2022YFD2100304）
国家自然科学基金资助项目(51705201) （51705201）
