TRAINING OF YOLOV5 NEURAL NETWORK FOR PEAR DETECTION IN ORCHARD

Authors

  • Artis Fribergs Rēzeknes Tehnoloģiju akadēmija
  • Edmunds Lukaševics Rēzeknes Tehnoloģiju akadēmija
  • Guntis Lielbārdis Rēzeknes Tehnoloģiju akadēmija
  • Sergejs Kodors Zinātniskā darba vadītājs, Dr.sc.ing., Rēzeknes Tehnoloģiju akadēmija

DOI:

https://doi.org/10.17770/het2023.27.7371

Keywords:

artificial intelligence, fruits, object detection, precision horticulture, YOLOv5,

Abstract

A fruit-growing is an important branch of agriculture for various reasons. Fruits provide essential nutrients and vitamins to our diet, and they are also a significant source of income for fruit-growers. To improve the efficiency of fruit cultivation, we trained a pear detection neural network with YOLOv5 architecture using a dataset from the project lzp-2021/1-0134. The dataset contained 1273 photographs of pear trees with image sizes 640x640px. We had trained the neural network model YOLOv5m five times and achieved the best result equal to mAP@0.5 0.8 and mAP@0.5:0.95 0.43. The use of artificial intelligence in fruit cultivation can help to optimize the planning of fruit picking, contributing to the precision horticulture.

Downloads

Download data is not yet available.

References

MakeSense. https://www.makesense.ai, sk. 02.04.2023.

Ultralytics. How to train Custom Data with YoLoV5. https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data, sk. 02.04.2023.

NVIDIA. CUDA. https://developer.nvidia.com/cuda-toolkit, sk. 02.04.2023

Eléa Petton. Object detection: train YOLOv5 on a custom dataset. https://blog.ovhcloud.com/object-detection-train-yolov5-on-a-custom-dataset/, sk. 02.04.2023

Downloads

Published

2023-10-30