Diagnosis of Late Blight of Potato Leaves Based on Deep Learning Hyperspectral Images

  • Kexing Cao
Keywords: Deep Learning, Hyperspectral Image, Feature Extraction, Late Blight Identification


Since late blight can cause devastating disasters to potatoes, the characteristics of hyperspectral imagery of
potato leaves stressed by late blight were studied. The aim is to explore the correlation between the
characteristics of the hyperspectral image of potato leaves and the degree of late blight in order to achieve
an accurate, fast and non-destructive diagnosis of late blight. Hyperspectral remote sensing technology has
received more and more attention in recent years. This is because hyperspectral images contain a large
amount of spectral information, spatial information, and radiant energy. classification. However, as the
dimensions of remote sensing data continue to increase, problems such as large amounts of image data,
data redundancy, and spectral correlation need to be resolved. How to extract the deep-level features of the
data often determines the quality of the classification results. With the development of deep learning
technology, feature extraction of hyperspectral images based on deep neural networks is a research hotspot
in the field of machine learning in recent years. In this paper, a deep neural network suitable for
hyperspectral images is constructed, and the network parameters are trained to achieve the optimal by
fine-tuning. Experiments show that the model can extract more abstract and easier classification features,
which can better explore the correlation between the hyperspectral image features of potato leaves and the
degree of late blight. The overall accuracy rate reached 99.34%. It is shown that the hyperspectral image
diagnosis technology based on deep learning can effectively distinguish the degree of potato disease under
the stress of late blight.