Related Experiment Video
Updated: Jun 28, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
529
Multiscale apple recognition method based on improved CenterNet
1College of Mechanical and Electrical Engineering, Hainan Vocational University of Science and Technology, Haikou, 571126, Hainan, China.
Heliyon
|April 18, 2024
Summary
This study introduces a fast CenterNet apple recognition method to improve real-time detection for apple-picking robots in complex environments. The enhanced YoloV5 model achieved high accuracy, even in challenging dark or occluded conditions.
Area of Science:
- Robotics
- Computer Vision
- Agricultural Technology
Background:
- Traditional apple-picking robots struggle with real-time apple detection in complex agricultural settings.
- Accurate fruit identification is crucial for efficient automated harvesting.
Purpose of the Study:
- To develop an efficient and accurate apple recognition method for dense scenes, enhancing robotic harvesting capabilities.
- To improve the real-time detection of multiple apple targets in challenging environments.
Main Methods:
- A fast CenterNet apple recognition method was proposed, utilizing a resnet-44 fully convolutional network, region of interest network (RPN), and region of interest (ROI).
- An improved YoloV5 network model was employed for experimental validation.
Main Results:
- The improved YoloV5 model achieved high recognition accuracies of 94.1% and 95.8% for apples in nighttime conditions.
- The method demonstrated improved recognition of occluded and dark-light features, showing increased robustness on actual datasets.
Conclusions:
- The proposed CenterNet method, particularly with the improved YoloV5 model, significantly enhances apple detection efficiency and accuracy for robotic applications.
- The developed model is robust and effective in complex, low-light, and occluded scenarios, paving the way for more advanced automated harvesting.
