Intelligent recommendation system for apple orchard protection in the Kabardino-Balkarian Republic

A. Z. Temrokov ; K. Ch. Bzhikhatlov

News of the Kabardin-Balkar scientific center of RAS, Tome 27 (2025) no. 2, pp. 23-36

Résumé

One of the important areas of agriculture is fruit gardening, in particular, intensive apple orchards make a significant contribution to the agricultural sector of the Kabardino-Balkarian Republic. At the same time, to preserve the harvest, it is necessary to ensure timely detection and elimination of threats associated with apple diseases and pests. Given the shortage of specialized specialists, the task of developing an automated system for recognizing diseases and pests of apple orchards becomes urgent. For this purpose, the study set the goal of developing and assessing the applicability of an intelligent recommendation system for the protection of apple orchards in the KBR. This article describes the concept and presents the results of the development of a system for monitoring the condition of apple orchards, designed to identify diseases and pests on trees, as well as select the most appropriate plant protection plan depending on the location of the orchard. The program is a web application created on the basis of the FastAPI, Vue.js frameworks and a neural network, responsible for recognizing pests and diseases of apple trees from a photograph and drawing up an optimal plan for their treatment. The results of training a neural network on a prepared sample of photographs of healthy and infected apples are presented. Various models were used as a basis for the neural network: Roboflow 3.0, RF-DETR, YOLO v11 and YOLO v12. The developed service will allow diagnosing apple tree diseases with minimal time delays, as well as ensuring the selection of protection methods, if necessary, which will reduce the risks of crop loss by gardeners. As a result of testing the model, the Roboflow 3.0 model achieved the best indicators: mAP was 91.0%, precision 97.5%, and recall 88.5%, which indicates the applicability of the approach. In order to expand the list of recognizable threats and improve accuracy, it is planned to collect additional photographic materials in the republic's orchards, including photographs of leaves and trunks of apple trees, and further testing with the participation of gardeners of the republic.

Détail

Keywords: image recognition, apple tree, apple diseases, recommendation system, machine learning
Mots-clés : internet service