Improving methods for construction of neural networks as a tool for environmentally friendly sunflower protection techniques

The object of this study is the processes of sunflower disease identification using neural networks and their impact on the efficiency and environmental sustainability of biological protection methods. The research addresses the task of improving the diagnosing accuracy of sunflower disease under conditions of limited real-world data. Specifically, this paper focuses on finding ways to enhance neural network design methods in data-scarce environments to improve the environmental sustainability of sunflower protection methods. A key feature of the results is the ability of the synthetic data integration algorithm to achieve high accuracy even with a limited amount of real data, which provides a significant advantage over conventional methods requiring large volumes of information. The application of mathematical modeling and Few-shot learning algorithms, combined with Generative Adversarial Networks (GANs) for generating synthetic images, improved diagnostic accuracy to 93–95 %, even with small datasets. This was achieved due to the model's high generalization capacity, trained on diverse synthetic data that accounted for varying field conditions. The findings make it possible to effectively apply biological protection methods by optimizing disease diagnosis based on mathematical modeling of the relationships between environmental conditions and biological agents. The practical significance of the results is the ability for agricultural practitioners to employ innovative diagnostic methods to enhance sunflower yield and reduce dependence on chemical protection agents. The proposed approaches contribute to the implementation of international environmental standards and could be integrated into agricultural decarbonization programs. The implementation of biological protection methods reduces environmental risks, saves resources, and maintains agroecosystem productivity