All publications from your selection

Download RAW file:
DOI:
Description: Heated and immature canola kernels caused by excessive drying and frost damage are undesired by grain buyers due to low oil yield, hence its presence above a certain threshold brings down their market value. These damages are identified by analyzing the endosperm colour which is different for healthy and damaged kernels. The current method employed for determining the damage is by visually examining the endosperm colour, and then counting the number of damaged seeds, which is laborious, time-consuming, and highly prone to both human and sampling error. This study proposes a three-stage deep learning based computer vision technique to detect, track, and count the damaged canola seeds in real time. The detection task was done using a YOLOv7 object detection network to localize and classify immature and heated seeds, which was trained on an annotated dataset containing 1500 instances. The detection weights obtained from the network were then inputted into a multiple object tracker to track the detections frame by frame and generate unique IDs which were then used to count the number of defects. The combined detector and tracker model counted the number of damaged kernels in an unseen video with good accuracy and an average inference speed of 24 FPS. Thus, the proposed system can be readily deployed in an edge device for accurate and real-time grading of canola kernels by grain buyers.

Keywords: Canola Quality, Grading, Object Detection, Computer Vision, Multi-object tracking
Résumé:
Mots-clés:
Citation:
Volume:
Issue:
Pages -
Contributor:
Date: 2023-07-23
Technical field:
Conference name: CSBE/SCGAB 2023 Annual Conference, Lethbridge, Alberta, 23-26 July 2023.
Session name:

Other information:
Type: Presentation
Format:
Publication type: Text.Abstract
Source:
Relation:
Coverage: North_America
Language 1: en
Language 2:
Rights: Canadian Society for Bioengineering
Notes:
Other files: