基于彩色图像的烟叶智能分选研究  

Research on Intelligent Grading Technology of Tobacco Leaves Based on Color Images

作　　者：位辉琴[1] 张长安 李彦周[1] 王孝奇 刘江豫[1] 贺凡 张相辉 徐世臻 WEI Huiqin;ZHANG Chang'an;LI Yanzhou;WANG Xiaoqi;LIU Jiangyu;HE Fan;ZHANG Xianghuil;XU Shizhen(Tianchang Internationgal Tobacco Co.,Ltd.,Xuchang,He'nan 461000,China;Shanghai Qidi Ruishi Intelligent Technology Co.,Ltd.,Shanghai 201210,China)

机构地区：[1]天昌国际烟草有限公司,河南许昌461000 [2]上海启迪睿视智能科技有限公司,上海201210

出　　处：《农产品加工》2024年第16期71-76,80,共7页Farm Products Processing

摘　　要：为解决打叶复烤企业人工烟叶分选准确率和效率低等问题,通过传统图像处理与神经网络相结合的方法对彩色图像进行处理,从而实现智能选叶。在分选预处理部分,采用语义分割分类网络进行烟叶状态和非主等级类检测提高分选时效性,在烟叶主等级分类部分,采用多维度的正面、背面、叶脉和叶尖不同特征建模,提高模型的鲁棒性和泛化能力。结果表明,单张图像的处理时间小于200 ms,青杂类分类准确率达到96.5%,主等级准确率达到82.1%。基于彩色图像的烟叶智能分选技术提高了打叶复烤烟叶分选效率和准确率,降低了劳动强度,加快了复烤企业分选智能化进程。To solve the problems of low accuracy and efficiency of manual grading of tobacco leaves in threshing and redrying enterprises.In order to realize intelligent grading through dealing with color images by the combination of traditional image processing and neural network method to.In the preconditioning of grading preprocessing,the semantic segmentation classification network was used to detect tobacco leaf status and non-main grade to improve the grading timeliness.In the main grade classification of the tobacco leaves,Building the model through the multi-dimensional different features of the front,back,vein,and tip of the tobacco leaves to enhance the robustness and generalization ability of the model.The results showed that:The processing time of a single image was less than 200 ms,and the grading accuracy of green and variegated leaf reached 96.5%,the grading accuracy of the main grade reached 82.1%.The intelligent grading technology based on color images improved the grading efficiency and accuracy,reduced the labor intensity,and accelerated the intelligent grading process of threshing and redrying enterprises.

关 键 词：复烤企业 智能分选 彩色图像 烟叶状态 烟叶表达特征 

分 类 号：TS44[农业科学—烟草工业]