摘要
针对当下在自动化和智能化的生产生活中增强生产者和消费者购买米种的透明度问题,方便消费者自行判断购买米种信息,由于判断人工识别的效率低、识别能力差,提出了一种适用于在线颗粒分割,并基于改进Convnext网络实现大米种类检测的方法。为实现自动化在线米粒的种类和品质分类应用,采用Cmos类相机对多种米粒的外观信息进行了采集,包括12000张图片6类米种;对采集的米粒图像采用匹配点区域选取和分割点匹配策略,对4种情况下的黏结米粒图像进行了外观黏结分割;在此基础上运用Convnext对采集稻米图片进行种类分类。为实现更好效果的分类,在Convnext的基础上引入多尺度特征融合机制和PSC注意力机制得到了Convnext-Mix-PSC(ConvMP)模型,改造convnext block模块完成位置和通道和多尺度特征的混合,再加入PSC注意力机制增强模型的自适应地集成重要的局部特征和全局特征,以单米粒图像为输入对象成功分类米粒的品种并且将准确率由90.59%提高到94.98%,验证了其在面向颗粒作物的现代农业信息化和食品安全检测中有较好效果。
To enhance the transparency of rice purchase by producers and consumers in the current automated and intelligent production and life and facilitate consumers to judge rice purchase information,a method for online particle segmentation and rice type detection based on improved Convnext network was proposed due to the low efficiency and poor recognition ability of artificial recognition.To realize automatic online classification of rice types and quality,Cmos cameras were used to collect the appearance information of a variety of rice grains,including 12000 pictures and 6 types of rice.Matching point region selection and the segmentation point matching strategy were used for the collected rice grain images,the appearance bonding segmentation of the rice grain images under four conditions was carried out.On this basis,Convnext was used to classify the collected rice pictures.To achieve better classification effects,the Convnext-Mix-PSC(ConvMP)model was obtained by introducing multi-scale feature fusion mechanism and PSC attention mechanism on the basis of Convnext,and the convnext block module was reformed to complete the mixing of location,channel and multi-scale features.Furthermore,the enhanced PSC attention mechanism model was added to integrate important local features and global features,rice varieties were successfully classified with a single rice grain image as the input object,and the accuracy was improved from 90.59%to 94.98%,verifying that the model had good effects in modern agricultural informatization and food safety detection for grain crops.
作者
黄文杰
刘珏
方焯
苗青
Huang Wenjie;Liu Jue;Fang Chao;Miao Qing(School of Electrical and Electronic Engineering,Wuhan Polytechnic University,Wuhan 430023)
出处
《中国粮油学报》
北大核心
2025年第5期169-179,共11页
Journal of the Chinese Cereals and Oils Association
基金
湖北省农机装备补短板核心技术应用攻关项目(HBSNYT202222)。
