Morphological feature assessment of Larimichthys crocea for quality grading: a computer vision approach

文献类型： 外文期刊

第一作者： Feng, Dejun

作者： Feng, Dejun;Liu, Hang;Gui, Fukun;Wu, Lianhui;Ying, Xiaoguo;Qu, Xiaoyu;Gao, Yang;Zhou, Chao

作者机构：

关键词： Large yellow croaker; Morphological feature detection; Grading evaluation system; Pectoral fin detection; Fin-to-eye length

期刊名称：AQUACULTURE INTERNATIONAL （ 影响因子：2.4； 五年影响因子：2.7 ）

ISSN： 0967-6120

年卷期： 2025 年 33 卷 5 期

页码：

收录情况： SCI

摘要： The large yellow croaker (Larimichthys crocea) stands as China's predominant marine aquaculture species economically, with annual yields surpassing 250,000 tons. While production volumes remain robust, market valuation demonstrates a strong correlation with morphological characteristics under equivalent weight conditions-particularly body shape, caudal fin length, and pectoral fin dimensions. Specimens exhibiting superior elongation metrics typically achieve premium market prices, often multiples above standard specimens, thereby necessitating precise morphological feature quantification in commercial aquaculture operations. Current manual assessments using vernier calipers and mechanical scales face three critical limitations: (1) time- and labor-intensive workflows, (2) operator-dependent subjectivity, and (3) measurement inaccuracies incompatible with rapid quality standardization. A persistent challenge in automated systems is pectoral fin detection failure due to fin transparency, leading to systematic metric exclusion. To address these issues, we developed a YOLMA (YOLO, MobileNet, AFPN) network that calculates two critical morphological indices: BL/BW (the body length/body width) and PFL/PEL (pectoral fin length/pectoral fin base to eye center length), where enhanced PFL/PEL correlates with swimming capacity and flesh quality. Technical implementations include the following: (1) an asymptotic feature pyramid network (AFPN) is integrated into the original YOLOv8-seg (You Only Look Once version 8 segments) network to improve pectoral fin detection via enhanced feature fusion; (2) the MobileNetV3 is used to replace the backbone network for lightweight deployment on embedded/mobile devices; and (3) a region-specific morphological detection algorithm is designed to optimize accuracy. Experimental results demonstrate a 2.47% average relative error for BL/BW measurements and 5.34% for PFL/PEL, with an average detection speed of 14.7 ms per specimen. This study represents the first successful implementation of automated PFL/PEL ratio quantification in large yellow croaker assessment. The achieved sub-6% error margin in fin measurements satisfies commercial grading requirements and addresses a critical gap in marine aquaculture automation while establishing quantitative benchmarks for morphological feature assessment in commercial fisheries.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序