基于改进YOLOv5s的水果新鲜度检测算法研究  

Research on fruit freshness detection algorithm based on improved YOLOv5s

作　　者：孙健飞 王莉[1] 王建鹏 SUN Jianfei;WANG Li;WANG Jianpeng(School of Electrical Engineering,Henan University of Technology,Zhengzhou 450001,China)

机构地区：[1]河南工业大学电气工程学院,河南郑州450001

出　　处：《现代电子技术》2024年第22期37-43,共7页Modern Electronics Technique

基　　金：河南省科技攻关项目(222102110160);河南工业大学创新基金支持计划专项资助(2022ZKCJ03)。

摘　　要：水果新鲜度分级在食品行业中有着重要作用。新鲜度是衡量水果质量的重要标准,直接影响到消费者的身体健康和购买欲望。由于水果颜色、纹理和外部环境变化(如阴影、照明和复杂背景)的相似性,使用机器视觉对水果进行自动识别和分类是具有挑战性的。文章提出一种基于改进YOLOv5s网络模型的多类水果分级方法。首先,引入DIoU-NMS算法,考虑预测框与真实框之间的重叠率以及中心点距离,回归精度得到提高;其次,利用K-means算法对初始锚框进行调整;最后,在主干网络Backbone中嵌入CAM,加强网络的特征提取能力。试验结果表明:改进后的YOLOv5s水果新鲜度检测算法平均检测一张图像耗时为0.028 s,且其mAP达到96.6%,比原来YOLOv5s模型提升了2.4%。所提方法为水果新鲜度检测提供一种高性能的解决方案,并能够以较高的准确率对多类水果进行分级与定位。The freshness grading of fruits plays an important role in the food industry.Freshness is the basic standard for measuring fruit quality,which directly affects consumers' physical health and purchasing desire.Due to the similarity in fruit color,texture,and external environmental changes(such as shadows,lighting,and complex backgrounds),using machine vision for automatic recognition and classification of fruits is challenging.A multi-class fruit grading method based on an improved YOLOv5s network model is proposed.The DIoU-NMS(distance IoU non-maximum suppression) algorithm is introduced improve regression accuracy by taking into account the overlap rate between the predicted box and the true and center point distance.The K-means algorithm is used to adjust the initial anchor box.CAM(context augmentation module) is embedded in the Backbone network to strengthen the feature extraction capability of the network.The experimental results indicate that the improved YOLOv5s fruit freshness detection algorithm takes 0.028 s to detect an image on average,and its mAP can reach 96.6%,which is 2.4% higher than the original YOLOv5s model.This research provides a high performance solution for fruit freshness detection,which can grade and locate many kinds of fruits with high accuracy.

关 键 词：水果新鲜度检测 YOLOv5s 深度学习 DIoU-NMS K-MEANS CAM 

分 类 号：TN911.23-34[电子电信—通信与信息系统]