As the agricultural landscape evolves, so does the need for innovative approaches to tackle persistent challenges. One such challenge is the effective management of viral diseases in crops, such as the Virus Yellows (VY) complex in sugar beets. This aphid-transmitted disease, primarily caused by beet mild yellowing virus (BMYV) and beet chlorosis virus (BChV), has resurged following the ban on neonicotinoid insecticides in the European Union. To address this, researchers are exploring cutting-edge technologies like unmanned aerial vehicles (UAVs) and convolutional neural networks (CNNs) for automated disease phenotyping.
A team of scientists from the Institute of Sugar Beet Research embarked on a study to assess whether high-resolution aerial imagery combined with CNNs could accurately score disease incidence in sugar beet fields. Traditional methods rely heavily on visual assessments by human experts, which are both time-consuming and subjective. This study aimed to determine if aerial multispectral and RGB (Red, Green, Blue) images could provide a more efficient and objective alternative.
The study utilised data from variety trials conducted in 2021 and 2022. The 2021 data was used to train and validate a CNN model on five different sugar beet cultivars, while the 2022 data, involving two different cultivars, tested the model’s real-world applicability. An interesting approach in this research was the use of transformed image features instead of raw spectral bands, which was found to improve the generalization capability of the CNN models.
Here are the key findings of this research:
- Model Performance: The CNN model trained on RGB images with transformed features achieved a root mean square error (RMSE) of 11.45%, indicating a strong correlation between the model’s scores and those of human experts. This error rate suggests however that while UAV and CNN technologies hold significant promise, they are not yet ready to completely replace human expertise.
- RGB vs. Multispectral Imaging: The study revealed that for diseases with subtle symptoms, such as BChV, multispectral imaging can detect changes not visible to the naked eye. However, for more pronounced symptoms like those of BMYV, RGB imaging performed equally well, challenging the necessity of more complex multispectral systems in such cases.
- Practical Implications: UAVs equipped with high-resolution cameras offer numerous benefits, including efficiency, objectivity, cost-effectiveness, and consistency in disease scoring. However, the complete replacement of human raters by UAVs requires careful consideration, especially for genotypes with subtle disease variations.
This research underscores the potential of UAVs and CNNs as powerful tools in modern agriculture. By automating disease phenotyping, these technologies can streamline the assessment process, reduce human subjectivity, and enhance breeding programs aimed at developing disease-resistant crop varieties. However, the journey towards fully autonomous disease scoring systems continues, with human expertise remaining crucial for validation and fine-tuning. For those interested in exploring this exciting field, the future is bright with opportunities to make impactful discoveries that will help feed the world.
Nathan Okole, Facundo R. Ispizua Yamati, Roxana Hossain, Mark Varrelmann, Anne-Katrin Mahlein and Rene HJ Heim published this work in Plant Pathology Journal:
TITLE IMAGE: More than 120 plots were scored for disease incidence by three experts and with images collected by a UAV-mounted camera. The scatter plot shows that the score obtained from the experts and the UAV, agrees for most plots.
