玉米霉变粒优势真菌分析及机器视觉识别技术研究  

Analysis of Dominant Fungi in Moldy Maize Kernels and Research on Machine Vision Identification Technology

作　　者：傅子恒 唐芳 纪立波 张海洋 田琳 雷雨晴 祁智慧 Fu Ziheng;Tang Fang;Ji Libo;Zhang Haiyang;Tian Lin;Lei Yuqing;Qi Zhihui(Academy of National Food and Strategic Reserves Administration,Beijing 100037;Key Laboratory for Northern Urban Agriculture of Ministry of Agriculture and Rural Affairs,Beijing University of Agriculture,Beijing 102206;National Engineering Research Center of Grain Storage and Logistics,Beijing 102209;China Grain Reserves Liaoning Quality Inspection Center Co.,Ltd.,Liaoning 110031)

机构地区：[1]国家粮食和物资储备局科学研究院,北京100037 [2]北京农学院农业农村部华北都市农业重点实验室,北京102206 [3]粮食储运国家工程研究中心,北京102209 [4]中储粮辽宁质检中心有限公司,沈阳110031

出　　处：《中国粮油学报》2025年第2期203-210,共8页Journal of the Chinese Cereals and Oils Association

基　　金：国家粮食和物资储备局科学研究院专项(JY2309)。

摘　　要：感染真菌的玉米籽粒发生霉变,会影响玉米质量,有些真菌可产生毒素,引发人畜中毒等食品安全问题。因此,研究玉米霉变识别技术,特别是产毒真菌的识别,为玉米真菌及毒素污染的防控处置提供参考。本研究对辽宁地区采集的不同颜色玉米霉变粒进行真菌分离鉴定,附生与内生真菌的优势菌群均为镰孢属(Fusarium)和木霉属(Trichoderma)。通过人眼识别对不同颜色玉米霉变粒进行分类,初步确定霉变粒颜色与优势内生菌及毒素污染水平具有关联关系。运用机器视觉识别技术的2种算法对不同颜色玉米霉变粒图像进行训练识别。K-means算法只能识别出玉米籽粒是否霉变;卷积神经网络不仅能区分是否霉变,对于霉变粒中优势菌属镰孢属和木霉属也可有效区分,进而可对毒素污染水平判定提供参考。Maize kernels infected with fungi will become moldy,affecting the quality of maize.Some fungi can produce toxins,causing food safety problems,such as poisoning of humans and animals.Therefore,studying maize mildew identification technology,especially the identification of toxin-producing fungi,can provide reference for the prevention and control of maize fungi and toxin pollution.In this study,fungi were isolated and identified from moldy maize grains of different colors collected in Liaoning.The dominant flora of epiphytic and endophytic fungi was both Fusarium and Trichoderma.By classifying moldy maize kernels of different colors through human eye recognition,it was initially determined that the color of moldy kernels was related to the dominant endophyte and toxin contamination levels.Two algorithms of machine vision recognition technology were used to train and identify images of moldy maize kernels of different colors.The K-means algorithm can only identify whether maize kernels are mildewed;the convolutional neural network can not only distinguish that whether maize kernels were mildewed or not,but can also effectively distinguish the dominant genera Fusarium and Trichoderma in mildewed kernels,thereby determining the level of toxin contamination for reference.

关 键 词：玉米 优势真菌 霉变籽粒 机器视觉技术 卷积神经网络 

分 类 号：TS21[轻工技术与工程—粮食、油脂及植物蛋白工程]