Related Experiment Video
Updated: Jun 29, 2026
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.5K
Towards a Machine Vision-Based Yield Monitor for the Counting and Quality Mapping of Shallots.
Amanda A Boatswain Jacques1, Viacheslav I Adamchuk1, Jaesung Park1
1Precision Agriculture and Sensor Systems Laboratory (PASS), Department of Bioresource Engineering, McGill University, Sainte-Anne-de-Bellevue, QC, Canada.
Frontiers in Robotics and AI
|May 28, 2021
Summary
A new machine vision system accurately counts and sizes shallots during harvest, providing real-time yield mapping for specialty crops. This technology aids in improving harvest management and increasing producer returns.
Area of Science:
- Agricultural Engineering
- Computer Vision
- Horticulture
Background:
- Specialty crops, unlike field crops, demand more resources and are sensitive to environmental changes, necessitating precise quality and quantity assessments during harvest.
- Existing computer and machine vision technologies excel in fruit crop analysis but lack detailed yield mapping for vegetable crops.
- Accurate yield data is crucial for optimizing management practices and maximizing financial returns for specialty crop producers.
Purpose of the Study:
- To develop and evaluate a machine vision-based yield monitor for in-situ size categorization and continuous counting of shallots during harvesting.
- To create a software system integrating video logging and global navigation satellite system (GNSS) for real-time data collection and analysis.
- To assess the system's precision in shallot detection and classification into different size categories.
Main Methods:
- A machine vision system utilizing an RGB camera and Python-based software was employed for real-time data collection under natural sunlight.
- Shallots were segmented using Watershed segmentation, detected on a conveyor, and classified by size using computer vision algorithms.
- The system integrated a video logger and GNSS for continuous data acquisition and yield mapping capabilities.
Main Results:
- The system achieved a 76% precision in detecting shallots within a dataset subsample.
- Size classification performance varied, with the highest accuracy for the large size category (73%), followed by small (59%), and medium (44%).
- Occasional vegetable occlusion and inconsistent lighting conditions were identified as primary factors limiting system performance.
Conclusions:
- The developed machine vision system offers a novel and modular approach for real-time yield monitoring and mapping of horticultural crops, particularly small vegetable crops.
- Despite challenges with occlusion and lighting, the prototype demonstrates significant potential for improving harvest information for producers.
- Further enhancements to the system are expected to improve accuracy and broaden its applicability in precision agriculture.
