Plant diseases pose a significant threat to countries around the world due to their economic burden and impact on food security. Healthy crops sustain the livelihoods of millions of people, and accurate diagnosis of plant diseases enables timely interventions to ensure adequate crop production while minimizing yield loss. becomes possible. Traditional approaches to disease recognition typically follow two paths. The first relies on crop inspection by trained professionals, the second leverages neural networks and image processing. However, both options have limitations. Trained experts provide opinions following error-prone and time-consuming manual inspection, whereas traditional image processing methods can only extract superficial information and make more accurate predictions. requires prerequisite training. This means that it is difficult to make consistent predictions as information complexity increases.
In this regard, neural networks — collections of algorithms that detect underlying relationships in data — have shown promising results in plant disease classification. The caveat is the lack of suitable training data. Good data collection is possible in a controlled environment, but not trivial in the real world. In the field, diseases may be rare or not easily observable. In addition, diseased samples may have complex backgrounds, different shapes, and occlusions. Neural network-based disease classifiers also have limited ability to apply knowledge to new datasets after training.
Now, the research team new and relatively simple A neural network called multi-representative subdomain adaptive network with uncertainty regularization for interspecific plant disease classification (MSUN) accurately classifies plant diseases in the natural environment. To do so, we applied a transfer learning technique called unsupervised domain adaptation (UDA) to smooth the plant disease identification process. The research was led by his Xijian Fan associate professor at Nanjing Forestry University, plant phenomicsThis paper was published online on March 28, 2023.
Huang, who is also the corresponding author of the study, explains: “UDA allowed the model to apply what it learned during training to another unannotated dataset. We trained MSUN to classify plant diseases in a controlled laboratory environment. UDA can now be used to classify plant diseases in complex field environments. “
The team’s approach to leveraging UDA for plant disease classification represents a paradigm shift that overcomes the shortcomings of current UDA-based approaches. First, images collected in the field are complex. There are some foliage, the camera shooting angle is weird, and it may be blurry. UDA-based classifiers are expected to process this confounding information for accurate disease classification. Second, these classifiers fail to make predictions when tackling plants suffering from different disease states, infections at different time points, or multiple sites. Third, classifiers face a significant challenge when similar disease symptoms can occur. This occurs when multiple pathogens infect a single plant species or when a single pathogen infects multiple plant species.
“MSUN is a more capable disease classifier at learning the global structure of plant disease features. It also gets more detail from the information it receives.” Mr. Huang said of the advantages of the new method: This study found that MSUN is not hampered by discrepancies that occur when the same information is collected in controlled environments and field settings. Importantly, this group validated the accuracy of MSUN’s disease classification using multiple complex plant disease datasets. When tested using data from the PlantDoc, Plant-Pathology, Corn-Leaf-Diseases, and Tomato-Leaf-Diseases databases, MSUN outperforms current classifiers!
The group is optimistic about MSUN’s prospects given its ability to handle challenging data sets. They believe it can overcome the uncertainties inherent in current disease classifiers and will aid future plant pathology research by providing important insights into the problem of disease recognition. I’m here.
– This press release was issued by Nanjing Agricultural University
