中国农机化学报2026，Vol.47Issue(9)：75-82,8.DOI:10.13733/j.jcam.issn.2095-5553.2026.09.010
基于改进YOLOv8的芒果成熟度检测方法
Maturity detection of mango based on improved YOLOv8
摘要
Abstract
To enhance the recognition accuracy of mangoes at different maturity stages and ensure high-quality harvesting,this study proposed a mango maturity detection method based on the YOLOv8—GAEW model.The method addressed common challenges in mango mechanical harvesting,such as small target sizes,dense fruit clustering,and severe occlusions in natural environments.Built on the YOLOv8 framework,the proposed approach incorporated a GhostNet lightweight backbone to reduce computational complexity and enable model lightweight processing.An AIFI scale feature interaction mechanism was introduced to improve the extraction and fusion of mango feature information,thereby enhancing detection accuracy.Additionally,the ECA attention mechanism was incorporated to effectively capture inter-channel dependencies and strengthen feature representation.The WIoUv3 loss function was adopted to prioritize medium-quality anchor boxes,further improving localization performance.Experimental results indicated that,compared with the original YOLOv8 model,the proposed YOLOv8—GAEW model reduced the number of parameters and floating-point operations by 22.26%and 34.15%,respectively.Meanwhile,detection accuracy of the YOLOv8—GAEW model increased from 97.45%to 99.23%.关键词
芒果/成熟度检测/轻量化模型/注意力机制/损失函数Key words
mango/maturity detection/lightweight model/attention mechanism/loss function分类
农业科技引用本文复制引用
王世刚,吴超敏,陈耀宗,冉运奇..基于改进YOLOv8的芒果成熟度检测方法[J].中国农机化学报,2026,47(9):75-82,8.基金项目
广西自然科学基金联合专项(2025GXNSFHA069207,2025GXNSFHA069265) （2025GXNSFHA069207,2025GXNSFHA069265）
