基于YOLOv5的鲜烟叶成熟度识别模型研究

Recognition model of tobacco fresh leaf maturity based on YOLOv5

【背景和目的】烟叶成熟度的准确判定和适时采收是提高烟叶质量的关键,为提高烟叶成熟度判定的准确性。【方法】以翠碧一号(CB-1)的鲜烟叶为研究对象,采用手机拍摄上、中、下3个部位5个成熟度(M1~M5)烟叶图像,利用labelimg软件从原始的图像中获取目标烟叶图像二维坐标信息,通过轻量级网络You Only Look Once(YOLO)v5进行数据训练,构建烟叶5个成熟度识别模型。【结果】CB-1的上、中、下3个部位模型中的m AP值均达到0.9以上,平均准确率分别为93.6%,92.8%,95.2%。进一步将模型部署到云服务器、并配套开发了基于Android端的烟草成熟度智能识别应用程序,实现在大田环境下响应式的鲜烟叶成熟度等级判断。【结论】基于YOLOv5模型的智能识熟APP可有效、准确地判定鲜烟叶成熟度。本研究结果可为鲜烟叶成熟度的智能识别提供理论基础和技术支撑。

[Background and objective]Determining tobacco leaf maturity accurately and harvesting on time are important for improving tobacco leaf quality.The purpose of this study is to improve the classification accuracy of fresh tobacco leaf maturity.[Methods]Taking CB-1 as the research object,five maturity(M1-M5)tobacco leaf images of the upper,middle and lower positions were taken by using the mobile phone.Labelimg software was to obtain the two-dimensional coordinate information of the target tobacco leaves images from the original images,and five maturity recognition models of tobacco leaves were constructed after training based on the lightweight network You Only Look Once(YOLO)v5s.[Results]The results showed that the mAP values in the upper,middle and lower position of CB-1 were more than 0.9,with average accuracy reaching 93.6%,92.8%and 95.2%respectively.The model was further applied on the cloud server.An Android-based intelligent tobacco maturity recognition application program was also developed to realize the responsive examination of the maturity of the tobacco leaves in a field environment.[Conclusion]The proposed APP can accurately and effectively classify the maturity of fresh tobacco leaves,which provides a theoretical basis and technical support for the intelligent recognition of tobacco leaf maturity.