摘要
[目的]提高苹果缺陷和分类准确率。[方法]提出一种基于改进YOLOv7-tiny的苹果缺陷识别方法。设计了多角度图像采集系统,对苹果表面进行采样和增强;利用YOLOv7-tiny网络提取苹果特征;通过改进模糊C均值聚类(IFCM)算法对提取的特征进行降维压缩;采用改进浣熊优化算法(ICOA)自动优化YOLOv7模型的超参数。对比分析不同分辨率、批量大小下,所提方法与ResNet+FPN、YOLOv5s、PP-YOLOE等方法的苹果缺陷识别与分类性能。[结果]所提方法在样本分辨率224像素×224像素时检测准确率可达98.6%,召回率达97.9%,单张图像平均检测时间仅50 ms左右,显著优于所对比方法。[结论]该系统具备高精度和实时性,能够有效提高苹果分类效率和质量,对水果自动分拣具有重要工程意义。
[Objective]To improve the accuracy of apple defect identification and classification.[Methods]An apple defect identification method based on improved YOLOv7-tiny is proposed.Firstly,a multi-angle image acquisition system is designed to sample and enhance the surface of the apple.Then,the YOLOv7-tiny network is used to extract the features of the apple.The extracted features are dimensionally reduced and compressed with the improved fuzzy C-means clustering(IFCM)algorithm.Finally,the improved coati optimization algorithm(ICOA)is adopted to automatically optimize the hyperparameters of the YOLOv7 model.The proposed method is compared with other methods,such as ResNet+FPN,YOLOv5s,and PP-YOLOE,in terms of apple defect identification and classification performance under different resolutions and batch sizes.[Results]When the sample resolution is 224 pixels×224 pixels,the proposed method achieves the detection accuracy of 98.6%and the recall rate of 97.9%and takes only about 50 ms to detect a single image on average,outperforming the other methods.[Conclusion]This system has high precision and real-time performance and can effectively improve the classification efficiency and quality of apples,which is of great engineering significance for the automatic sorting of fruits.
作者
李俊
曹博涛
彭新东
LI Jun;CAO Botao;PENG Xindong(Guangdong Province Nanfang Technician College,Shaoguan,Guangdong 512023,China;Shaanxi University of Science&Technology,Xi'an,Shaanxi 710021,China;Shaoguan University,Shaoguan,Guangdong 512005,China)
出处
《食品与机械》
北大核心
2025年第8期100-108,共9页
Food and Machinery
基金
陕西省国际科技合作计划重点项目(编号:2020KWZ-015)
广东省教育和职业培训课题项目(编号:KT2023019)。
