Scientists have developed a method to automatically recognize plant diseases from a photograph. Reported by Nanjing Agricultural University.
Plant diseases cause great damage to both agriculture and wildlife. In the case of agricultural crops, diseases place a financial burden on producers and can lead to famine in poor countries. Traditionally, the main method of diagnosing a plant is self-examination by a specialist, but they cannot be found everywhere and constantly.
Now, Xijian Fan and colleagues have created the Multi-Representative Subdomain Adaptive Network with Uncertainty Regulation for the Interspecies Classification of Plant Diseases (MSUN) neural network. The main challenge of creating it was creating a database large enough for training. This is complicated by the fact that diseases can manifest differently in different species, and the appearance of plants can differ even within the same species. Therefore, the authors had to apply a non-standard method: they had to train the model on a dataset obtained in the laboratory and adapt the result of this training for many other sets.
“The UDA method allowed our model to apply what it learned during training to another dataset. We trained MSUN to classify plant diseases in a controlled laboratory environment, and with UDA it was able to classify plant diseases under harsh field conditions,” write the authors.
As a result, the model can recognize diseases even in blurry photos from non-standard angles, as well as identify situations where the plant suffers from several problems at once.
ancient biologists can Treats memory impairment in mice with Alzheimer’s disease.