Pruning Points Detection of Sweet Pepper Plants Using 3D Point Clouds and Semantic Segmentation Neural Network

Area of Science:

• Agricultural Robotics
• Computer Vision
• Smart Farming Technologies

Background:

• Agricultural automation is crucial for increasing labor efficiency and productivity.
• Previous work focused on semantic segmentation for plant part identification.
• Accurate plant part detection is essential for robotic manipulation in agriculture.

Purpose of the Study:

• To develop an automated system for pruning sweet pepper plants in smart farms.
• To detect precise leaf pruning points in 3D space using 3D point clouds.
• To enable robot arms to execute pruning tasks accurately.

Main Methods:

• Utilized semantic segmentation neural networks to identify plant parts.
• Generated 3D point clouds of sweet pepper plants using ORB-SLAM3 and a LiDAR camera.
• Applied the ICP algorithm for point cloud registration and processing.
• Developed a method for detecting leaf pruning points in both 2D images and 3D space.
• Employed the PCL library for visualization of 3D point clouds and pruning points.

Main Results:

• Successfully created detailed 3D point clouds of sweet pepper plants with recognized plant parts.
• Demonstrated a robust method for detecting leaf pruning points in 3D space.
• Visualized 3D point clouds and identified pruning points using the PCL library.
• Experimental results confirmed the stability and correctness of the proposed method.

Conclusions:

• The developed method enables precise 3D localization of leaf pruning points for automated agricultural tasks.
• This approach significantly contributes to the advancement of robotic pruning in smart farming.
• The research validates the potential of 3D point cloud processing for enhancing agricultural automation.