基于迁移学习的烟草叶片病情害识别

Identification of Tobacco Leaf Disease Based on Transfer Learning

烟草叶片病情害种类繁多且病理复杂,仅仅依靠传统的人工识别方法比较困难。针对我国常见的烟草叶片病情害,本文运用迁移学习方法和微调(fine-tuning)方法通过训练深度卷积网实现病情害识别,主要研究运用构建的11类图片库,通过迁移学习实现深度卷积网分类模型,通过分析和比较两种模型得出了烟草叶片病情害识别采用Fine-tuning迁移学习模型效果更好的结论。

It is difficult to rely solely on traditional methods of artificial identification for tobacco leaves with a wide variety of diseases and complicated pathology.In view of the common tobacco leafdiseases in our country,this paper uses transfer learning method and fine tuning(fine-tuning)method to achieve disease identification by training deep convolution network,In this paper,we mainly study11types of image libraries,and implement deep convolution network classification model by transfer learning,through analysis and comparison of the two models,we get the conclusion that the Finetuning transfer learning model is better for tobacco leaf disease identification.