Polar-Net: Green fruit instance segmentation in complex orchard environment.

Area of Science:

• Robotics
• Computer Vision
• Agricultural Technology

Background:

• Automated fruit picking requires precise segmentation of homogeneous fruits, a significant challenge in complex orchard environments.
• Current picking robots face difficulties in accurately identifying and segmenting fruits with similar colors.

Purpose of the Study:

• To develop an efficient and accurate homo-chromatic fruit segmentation model for picking robots.
• To improve the picking efficiency of fruits with similar colors in challenging environments.

Main Methods:

• A novel homo-chromatic fruit segmentation model based on Polar-Net was proposed.
• The model utilizes Densely Connected Convolutional Networks (DenseNet) for feature extraction, Feature Pyramid Network (FPN), and Cross Feature Network (CFN) for multi-scale and cross-feature discrimination.
• Region Proposal Network (RPN) identifies regions of interest, and polar coordinate modeling is employed for instance segmentation by predicting contours.

Main Results:

• The proposed model significantly enhances the segmentation accuracy of homo-chromatic objects.
• Experimental results validate the model's simplicity and efficiency in complex environments.
• The method demonstrates improved performance for picking robots and offers a reference for segmenting other fruits and vegetables.

Conclusions:

• The developed model effectively addresses the challenge of segmenting homogeneous fruits for robotic picking.
• This research provides a valuable reference for improving fruit and vegetable segmentation in agricultural robotics.
• The approach offers a simple yet efficient solution for enhancing automated harvesting systems.