Building Deep Neural Networks for solving Machine Learning Problems in Agricultural Production

In the tasks of agricultural production, it is necessary to identify unfavorable situations of agricultural farming that arise in the process of cultivating agricultural crops. These include soil erosion or salinization, damage from crop diseases, pests, and others. Timely and prompt identification of such situations is possible with the use of technical vision and methods of intellectual analysis and image processing. The most effective means of machine learning (ML) for such tasks are deep neural networks (DNN), primarily based on a parallel architecture containing convolutional layers of neurons. The purpose of the study was to build and study the effectiveness of DNN, which are used in intellectual land use tasks. The Python-based Google Collaboration cloud service, including ML libraries, was used as the DNN development environment.. When designing DNN, the features of the functioning of the CPU and GPU were taken into account. The results obtained make it possible to optimize the architecture and hyperparameters of DNN, as well as their training time. This approach increases the efficiency of the information and analytical complexes being developed to support the solution of various land use problems.