Abstract
<jats:p>The task of developing neural network models that provide high accuracy in detecting plant pests with minimal computational and time costs is relevant. Solving this problem will contribute to the development of digital technologies in the agricultural industry and will provide an opportunity for prompt and accurate detection of threats to agricultural crops, in particular, the identification of plant pests. The purpose of the work is to develop an intelli-gent system for identifying plant pests based on neural network technologies and study its performance.During the study, an object detection model based on the YOLOV5s architecture was implemented and analyzed. For training and validation of the model, a dataset was used, formed on the basis of open datasets hosted on the Roboflow platform. The final dataset includes 3766 annotated images, each of which contains at least one object belonging to one of 18 predefined classes.To increase the generalization ability of the model and expand the diversity of the train-ing dataset, the Albumentations augmentation library was used at the preprocessing stage. Training was performed using a stochastic gradient descent optimizer. A cosine sched-uler was used to control the learning rate. User interaction with the server part with an interface in the form of a chat bot was also implemented. The developed intelligent system provides two full-fledged operating modes: neural network inference mode (predict) and data collection mode (collect). Switching be-tween modes is carried out both via commands and automatically - based on the presence of a model in the file system of the server part of the web application. The results of testing con-firmed the stable operation of the server part and demonstrate full compliance with the stated functional requirements.</jats:p>