# Enhancement of the prediction of the openness of fresh-cut roses with an improved YOLOv8s model validated by an automatic Grading Machine

## TL;DR

This paper introduces an improved YOLOv8s model combined with an automatic grading machine to efficiently and accurately assess the openness of fresh-cut roses.

## Contribution

An improved YOLOv8s model with MobileNetV3 backbone and EMA module is proposed for fresh-cut rose openness grading.

## Key findings

- Replacing YOLOv8s backbone with MobileNetV3 reduces model complexity and computation.
- The EMA module and Wise-IoU loss improve detection accuracy and convergence speed.
- Field experiments showed grading errors of 6.5% to 12.6% compared to manual grading.

## Abstract

The openness grading of fresh-cut roses relies heavily on manual work, which can be inefficient and inconsistent.

In this study, an improved YOLOv8s model is proposed for openness grading in conjunction with a newly developed automatic grading machine for fresh-cut roses. The model identifies unopened inner petals and classifies openness into five levels: degree 1, degree 2, degree 3, degree 4, and deformity. To enhance detection accuracy while reducing the model complexity and computation, the backbone network of YOLOv8s is replaced by MobileNetV3. Additionally, an Efficient Multi-scale Attention (EMA) module is introduced to enhance focus on critical features, and a Wise-IoU loss function is incorporated to accelerate convergence.

Field experiments revealed that the openness predictions made by the automatic fresh-cut roses grader had errors of 6.9%, 9.1%, 10.0%, 6.5%, and 12.6%, respectively, compared to manual predictions.

Therefore, the improved YOLOv8s-F model effectively meets the requirements of fresh-cut rose openness grading.