基于生成式对抗网络的草莓白粉病病斑定位方法研究

Study on the Lesions Location Positioning Method of Strawberry Powdery Mildew Based on Generatire Adversarial Networks

卷积神经网络在解决物体定位问题时,经常需要使用物体级别标签的数据集对网络模型进行训练,这样的标签工作量极大,往往无法实现。本文在对CAM思想及基于CAM的多种弱监督学习的物体定位方法比较研究的基础上,设计了基于GAN方法的草莓白粉病病斑定位DAG网络模型。通过实验验证了DAG网络模型可以实现草莓白粉病病斑的定位。

Convolutional neural networks often need to use the dataset of object-level tags to train the network model when solving object location problems.Such tag workloads are often impossible to achieve.Based on the comparative study of CAM thought and CAM-based multi-weak supervised learning object localization methods,this paper designs a DAG network model based on GAN method for strawberry powdery mildew lesion location positioning.It is verified by experiments that the DAG network model can achieve the lesion location positioning of strawberry powdery mildew.