An accurate green fruits detection method based on optimized YOLOX-m.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Accurate fruit detection is crucial for modern agricultural automation, including harvesting and yield prediction.
• Complex orchard environments present significant challenges for existing fruit detection methods.
• Automated systems require robust algorithms for reliable green fruit recognition.

Purpose of the Study:

• To develop and evaluate an optimized object detection model for accurate green fruit identification in challenging orchard settings.
• To enhance the performance of fruit detection systems for applications in precision agriculture.
• To address the limitations of current methods in detecting fruits under varying environmental conditions.

Main Methods:

• An optimized YOLOX_m model was proposed, utilizing CSPDarkNet for feature extraction.
• A feature fusion pyramid network with an Atrous spatial pyramid pooling (ASPP) module was employed for enhanced multi-scale feature extraction.
• Varifocal loss was implemented to address class imbalance and improve detection precision.

Main Results:

• The optimized YOLOX_m model achieved an average precision (AP) of 64.3% for apples and 74.7% for persimmons.
• The proposed method demonstrated superior performance compared to other common detection models on apple and persimmon datasets.
• Significant improvements were observed in various performance metrics, indicating enhanced detection capabilities.

Conclusions:

• The optimized YOLOX_m model provides a robust and accurate solution for green fruit detection in complex agricultural environments.
• The method offers a valuable reference for the development of detection systems for other fruits and vegetables.
• This advancement contributes to the automation and efficiency of agricultural practices through improved computer vision techniques.