RAPID PROTOTYPING OF PEAR DETECTION NEURAL NETWORK WITH YOLO ARCHITECTURE IN PHOTOGRAPHS

Abstract

Fruit yield estimation and forecasting are essential processes for data-based decision-making in agribusiness to optimise fruit-growing and marketing operations. The yield forecasting is based on the application of historical data, which was collected in the result of periodic yield estimation. Meanwhile, the object detection methods and regression models are applied to calculate yield per tree. The application of powerful neural network architectures for rapid prototyping is a common approach of modern artificial intelligence engineering. Meanwhile, the most popular object detection solution is YOLO architecture. Our project team collected the dataset of fruiting pear tree photographs (Pear640) and trained YOLOv5m with mAP@0.5 95% and mAP@0.5:0.95 56%. The obtained results were compared with other YOLOv5-7.0 and YOLOv7 models and similar studies.