| Weeds detection in saffron fields using an improved YOLOv5 object detection model |
| Paper ID : 1153-NICAME1402 |
| Authors: |
|
roghaye shamloo, Abbas Rezaei Asl *, Alireza Soleimanipour Department of Biosystem Engineering, Gorgan University of Agricultural Sciences and Natural Resources, Gorgan, Iran |
| Abstract: |
| The careless application of pesticides and fertilizers can lead to severe environmental impacts to the agricultural ecosystem. By implementing digital agriculture and variable rate application systems, it is possible to manage various sections of an agricultural field with varying levels of pesticides and inputs. This approach can provide advantages in terms of reducing production cost and minimizing environmental concerns. The focus of this research was development of a model for detecting weeds and saffron plants in the field, with the aim of implementing a selective control system for weeds that grow in saffron fields. The YOLOv5 object detection model was used as the basis for the proposed algorithm. To reduce the model's parameters and improve its efficiency during both training and inference, the CBS and C3 modules from the YOLOv5s model were replaced with Ghost Bottleneck and C3Ghost modules. This modification resulted in a lighter network and increased the speed of image processing. In order to enhance the precision of the model, an attention mechanism layer called CoordAtt was incorporated. The findings revealed that the proposed model had 47% fewer parameters compared to the YOLOv5 version with similar network depth and width. Furthermore, the proposed model achieved 3.93% higher precision than the variant of the YOLOv5 algorithm that performed the best. Because of its lightweight design, the proposed algorithm is suitable for real-time detection of weeds in agricultural fields, enabling the development of selective control systems. |
| Keywords: |
| Saffron, Weeds control, Object detection, Deep learning, YOLOv5 |
| Status : Paper Accepted (Poster Presentation) |
