[Paper Review] Machine Vision-Based Crop-Load Estimation Using YOLOv8

Dawood Ahmed, Ranjan Sapkota+2 more|arXiv (Cornell University)|Apr 26, 2023

Smart Agriculture and AI17 citations

TL;DR

The paper presents a YOLOv8-based machine vision system using RGB-D data to identify apple tree trunks and branches, estimate branch diameter and limb cross-sectional area, and predict crop-load per branch for robotic pruning and thinning. It reports ground-truth-aligned errors and demonstrates feasibility in commercial orchards.

ABSTRACT

Labor shortages in fruit crop production have prompted the development of mechanized and automated machines as alternatives to labor-intensive orchard operations such as harvesting, pruning, and thinning. Agricultural robots capable of identifying tree canopy parts and estimating geometric and topological parameters, such as branch diameter, length, and angles, can optimize crop yields through automated pruning and thinning platforms. In this study, we proposed a machine vision system to estimate canopy parameters in apple orchards and determine an optimal number of fruit for individual branches, providing a foundation for robotic pruning, flower thinning, and fruitlet thinning to achieve desired yield and quality.Using color and depth information from an RGB-D sensor (Microsoft Azure Kinect DK), a YOLOv8-based instance segmentation technique was developed to identify trunks and branches of apple trees during the dormant season. Principal Component Analysis was applied to estimate branch diameter (used to calculate limb cross-sectional area, or LCSA) and orientation. The estimated branch diameter was utilized to calculate LCSA, which served as an input for crop-load estimation, with larger LCSA values indicating a higher potential fruit-bearing capacity.RMSE for branch diameter estimation was 2.08 mm, and for crop-load estimation, 3.95. Based on commercial apple orchard management practices, the target crop-load (number of fruit) for each segmented branch was estimated with a mean absolute error (MAE) of 2.99 (ground truth crop-load was 6 apples per LCSA). This study demonstrated a promising workflow with high performance in identifying trunks and branches of apple trees in dynamic commercial orchard environments and integrating farm management practices into automated decision-making.

Motivation & Objective

• Address labor shortages in fruit production by enabling automated pruning, thinning, and harvesting workflows.
• Develop a vision-based method to identify tree trunks/branches and estimate geometric parameters that influence fruit load.
• Create a workflow integrating sensor data and machine learning to inform crop-load decisions for automated systems.

Proposed method

• Use RGB-D data from Microsoft Azure Kinect DK for color and depth information.
• Apply YOLOv8-based instance segmentation to identify trunks and branches during the dormant season.
• Apply Principal Component Analysis to estimate branch diameter and orientation from segmented structures.
• Compute limb cross-sectional area (LCSA) from estimated branch diameter to serve as input for crop-load estimation.
• Estimate target crop-load per segmented branch based on LCSA, reporting MAE against ground truth.

Experimental results

Research questions

• RQ1 Can YOLOv8-based instance segmentation accurately identify trunks and branches in commercial apple orchards during the dormant season?
• RQ2 How well can branch diameter and orientation be estimated from RGB-D data and used to compute LCSA for crop-load estimation?
• RQ3 What is the accuracy (MAE) of the predicted crop-load per branch compared to ground truth?
• RQ4 Is the proposed workflow robust enough for integration with automated pruning and thinning platforms in real orchard settings?
• RQ5 What are the practical performance metrics (e.g., depth accuracy, segmentation performance) achievable in real-world environments?

Key findings

• The branch-diameter estimation error achieved a mean performance metric of 2.08 mm (for diameter estimation).
• Crop-load estimation per branch achieved a mean absolute error of 2.99 apples per branch where ground truth is 6 apples per LCSA.
• The approach enables high-performance identification of trunks and branches in dynamic commercial orchard environments.
• The workflow supports integration of farm management practices into automated decision-making for pruning and thinning.
• The system demonstrates feasibility of using RGB-D data and YOLOv8 for crop-load related decisions in practical settings.