进境原粮携带菊科和豆科杂草籽智能全维识别方法研究  

Study on intelligent full-dimensional identification of Asteraceae and Fabaceae weed seeds carried in imported raw grains

作　　者：刘新娇 张伟锋 吴影仁 卢小雨 汪莹 王伍 吴煜梓 章桂明 Liu Xinjiao;Zhang Weifeng;Wu Yingren;Lu Xiaoyu;Wang Ying;Wang Wu;Wu Yuzi;Zhang Guiming(Animal and Plant Inspection and Quarantine Technology Center of Shenzhen Customs District,Shenzhen 518045,China;Guangdong Penghai Forensic Science Institute;Shenzhen CIMC Intelligent Technology Co.,Ltd.)

机构地区：[1]深圳海关动植物检验检疫技术中心,广东深圳518045 [2]广东鹏海司法鉴定所 [3]深圳中集智能科技有限公司

出　　处：《植物检疫》2024年第4期19-25,共7页Plant Quarantine

基　　金：国家重点研发计划项目(2022YFF0608803)。

摘　　要：应用全维智能有害生物识别仪,建立基于卷积神经网络技术的进口原粮携带杂草籽形态特征模型,并将该仪器和模型应用于进口原粮携带的16种菊科杂草籽和18种豆科杂草籽的全维智能识别研究,将杂草籽放置在不同方位进行多次检测,分析了其检测准确性情况,结果表明:在已建模的30种物种中,22个物种识别正确率为100%;2个物种识别正确率为95%~99%;6个物种低于95%。而4种未建模的物种中,3种不能识别,1种误识别为其他物种。说明该仪器及开发的物种识别模型对进境原粮携带的杂草籽具有很好的识别效果。研究还发现,不同放置方位对识别率有微量影响,放在边缘位置可能出现不能识别的情况;而不同个体对识别正确率影响较大,识别正确率可能与标本个体表面特征是否完备相关。A full-dimensional intelligent identification instrument was applied to establish a species characteristic model of imported raw grain carrying weed seeds based on convolutional neural network technology,and the instrument and model were applied to the intelligent full-dimensional identification research of 16 kinds of weed seeds of Asteraceae family and 18 kinds of weed seeds of Fabaceae carried by imported raw grain. These weed seeds were placed in different directions of the instrument for multiple detection. The accuracy of the detection were analyzed. The results showed that the identification accuracy of 22 species was 100% among the 30 species modeled,2 species was 95%-99%. 6 species were below 95%. Of the 4 unmodeled species,three could not be identified and one was misidentified as another species. It shows that the instrument and the species identification model have a good effect on the identification of weed seeds. It is also found that the different place orientation has a slight influence on the recognition rate,and the edge position may not be recognized. However,different individuals have a great influence on the recognition rate,and the recognition accuracy may be related to whether the surface features of the specimen are complete.

关 键 词：进境原粮 杂草籽 智能识别 菊科 豆科 

分 类 号：S41[农业科学—植物保护]