基于改进YOLOv5稻米垩白检测的研究  

Research on Rice Chalkiness Detection Based on Improved YOLOv5

作　　者：喻伟 周劲 黄金良 Yu Wei;Zhou Jin;Huang Jinliang(School of Electrical and Electronic Engineering Wuhan Polytechnic University,Wuhan 430048)

机构地区：[1]武汉轻工大学电气与电子工程学院,武汉430048

出　　处：《中国粮油学报》2025年第2期41-48,共8页Journal of the Chinese Cereals and Oils Association

基　　金：湖北省重点研发计划项目(2023BBB110)。

摘　　要：稻米垩白是稻米不成熟的表现,这是反映稻米质量的标志之一。垩白不仅影响了稻米的外观品质,同时也影响稻米口感。传统的稻米垩白检测是通过人工分辨进行筛选,不仅筛选的误差高,而且效率也非常低下。针对效率低,误差高的问题,本研究提出了一种基于改进的YOLOv5稻米垩白检测算法,采用自建的一个稻米垩白检测数据集,利用其改进的YOLOv5算法进行训练检测。该算法基于YOLOv5模型,通过重组网络结构、改进损失函数、加入SE和ECA注意力机制,使原YOLOv5的检测精度得到了提升。实验表明,其改进的YOLOv5算法平均检测精确率可达91.9%。研究得出采用注意力机制叠加比单独加入注意力机制模块模型效果更好。Chalkiness in rice is a sign of its immaturity,serving as one of the indicators reflecting rice quality.Traditionally,the detection of chalkiness in rice relies on manual screening,not only highly error-prone,but also inefficient.To address these issues of low efficiency and high error rates,in this paper,an improved YOLOv5 rice chalkiness detection algorithm utilizing a self-constructed rice chalkiness detection dataset was proposed.The improved YOLOv5 algorithm was used for training and detection.Based on the YOLOv5 model,the algorithm improved the detection accuracy of the original YOLOv5 by reorganizing the network structure,modifying the loss function,and adding SE and ECA attention mechanisms.Experimental results indicated that the improved YOLOv5 algorithm achieved an average detection accuracy of 91.9%.In this paper,thought experiments,it was also concluded that the combination of attention mechanisms performed better than adding individual attention mechanism modules.

关 键 词：深度学习 垩白检测 YOLOv5 SE注意力机制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]