烟草叶部病害种类繁多,病理复杂,严重影响烟草产量及品质,烟草病害精准检测是烟草病害及时防治的前提.传统检测方式精准性差、效率低,基于深度学习的算法可提高烟草病害检测准确性.本文以5种较为常见的烟草病害(普通花叶病、黄瓜花叶病毒病、赤星病、烟草野火病、气候性斑点病)为研究对象,构建基于YOLOv3的烟草病害检测模型,实现烟草多类病害的精准快速检测.使用Darknet53特征网络提取烟叶病害特征并将不同尺度病害特征融合,并用K-means++算法对融合后特征进行分类和位置预测,通过非极大值抑制算法(NMS)去除冗余框,得到病害区域预测框.用田间实际采集的烟草病害数据集,对构建的YOLOv3病害检测模型与SSD(Single Shot multibox Detector)模型对比测试.结果表明,YOLOv3的mIoU为0.81,明显优于SSD的0.73,且YOLOv3模型的mAP为0.77,也高于SSD的0.69.本研究构建的YOLOv3烟草病害检测模型能有效定位烟叶病害区域,实现多类烟草病害的检测,为精准病害防治提供参考.
The dynamic forecast and diagnosis information systems of Tobacco pests and diseases was studied based on WebGIS development system,using J2EE and drawing support Struts and Hibernate development framework,with Oracle relational database for data storing.This paper focuses on the tobacco pests and diseases seeking and diagnosis system's construction,and introduces system model design,features modular design,Oracle Database.
The small-scaled house of flue-cured tobacco is made of Temperature & humidity sensor,the module of four control fans,the module of automatic or man-designed baking model and the model of number display and output etc.The equipment control model take three paragraph five steps art as control model,trough automatic control about eleven key points of temperature & humidity and time,realizing the automatic monitor and control of the baking process.The experimental results in honeycomb curing barn and vertical furnace curing barn of hot-winded cycle shoe that the furnace temperature / humidity and time in barn operate basically on the inputing parameters when baking.As the honeycomb curing barn heating delayed,The temperature / humidity existe 0.3 tolerances.
