摘要
针对传统玉米病害的识别方法过于依赖个人经验、出错概率高的问题,结合深度学习与计算机视觉技术,提出一种基于迁移学习和MobileNetV2模型的识别方法,能够较准确地识别常见的三类玉米病害:灰斑病、锈病、大斑病。与经典CNN网络模型LeNet对比后,发现该方法具有计算量少、准确率高的特点,适用于移动设备。该方法在Kaggle数据集上的平均识别率达到96.94%,模型大小只有8.95MB。
In order to solve the problem that traditional identification methods of maize diseases are too dependent on per⁃sonal experience and have high error probability,a recognition method based on transfer learning and MobileNetV2 model was pro⁃posed combining deep learning and computer vision technology,which can accurately identify three common maize diseases:gray spot,rust and Northern leaf blight.Compared with LeNet,it is found that this method has the characteristics of less computation and high accuracy,and is suitable for mobile devices.The average recognition rate of this method on The Kaggle dataset is 96.94%,and the model size is only 8.95 MB.
作者
叶名炀
张杰强
Ye Mingyang;Zhang jieqiang(College of Electronic Engineering(Artificial Intelligence),South China Agricultural University,Guangzhou 510642)
出处
《现代计算机》
2022年第11期46-50,共5页
Modern Computer
