基于深度学习的发霉花生识别技术  

Research on moulded peanut recognition technology based on deep learning

作　　者：王伟娜[1] 许世维 邓勤波 李博 WANG Weina;XU Shiwei;DENG Qinbo;LI Bo(Henan College of Transportation,Zhengzhou,Henan 451460,China;Xi'an University of Architecture and Technology,Xi'an,Shaanxi 710054,China;Shaanxi University of Science&Technology,Xi'an,Shaanxi 710021,China;Shaanxi Fengrun Intelligent Manufacturing Research Institute Co.,Ltd.,Xi'an,Shaanxi 712000,China)

机构地区：[1]河南交通职业技术学院,河南郑州451460 [2]西安建筑科技大学,陕西西安710054 [3]陕西科技大学,陕西西安710021 [4]陕西风润智能制造研究院有限公司,陕西西安712000

出　　处：《食品与机械》2023年第8期136-141,共6页Food and Machinery

基　　金：陕西省自然科学基础研究计划项目(编号:2020JQ-669);陕西省秦创原“科学家+工程师”队伍建设项目(编号:S2022-ZC-QCYK-0050)。

摘　　要：目的:快速、无损地识别发霉花生,提高发霉花生的识别效率。方法:采用光谱仪采集高光谱花生数据,利用深度学习技术识别霉变花生,建立Hypernet PRMF模型,并以Deeplab v3+、Segnet、Unet和Hypernet作为对照模型进行比较。将所提出的花生识别指数融合到高光谱图像中,作为数据特征预提取。同时将构建的多特征融合块集成到控制模型中以提高发霉花生识别效率。结果:所有模型的平均像素精度均超过了87%。Hypernet-PRMF模型的检测精度最高,达到90.35%,同时对于整个花生数据集,Hypernet-PRMF的错误识别率较低,可以有效识别图中所有的发霉花生。结论:基于深度学习所建立的Hypernet-PRMF模型具有较高的像素精度与检测精度,可有效识别发霉花生。Objective:To identify mouldy peanuts in a fast and non-destructive way and improve the identification efficiency.Methods:Collected hyperspectral peanut data using a spectrometer,identify moldy peanuts using deep learning technology,and established a Hypernet PRMF model,which was compared with Deeplab v3+,Segnet,Unet,and Hypernet as control models.Integrated the proposed peanut recognition index into hyperspectral images as data feature pre extraction.Simultaneously integrating the constructed multi feature fusion blocks into the control model to improve the recognition efficiency of moldy peanuts.Results:The average pixel accuracy of all models exceeded 87%.the Hypernet-PRMF model had the highest detection accuracy of 90.35%,while for the whole peanut dataset,Hypernet-PRMF had a low false recognition rate and could effectively identify all mouldy peanuts in the figure.Conclusion:The Hypernet-PRMF model built based on deep learning has high pixel accuracy and detection precision,which can effectively identify mouldy peanuts and provide a reference basis for the identification and detection of other mouldy food and other hyperspectral objects.

关 键 词：深度学习 花生 霉变 识别 多特征块 

分 类 号：TS207.3[轻工技术与工程—食品科学] TP18[轻工技术与工程—食品科学与工程] TP391.41[自动化与计算机技术—控制理论与控制工程]