Corn classification based on quality related to the bug-disease using YOLOv8

Mohammed Al-Asbahi, Monther Yousef Abdulwasea (2024) Corn classification based on quality related to the bug-disease using YOLOv8. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Corn classification based on quality related to the bug-disease using YOLOv8.pdf Download (2MB)

Abstract

Artificial intelligence (AI) breakthroughs have transformed crop monitoring and quality control techniques in the agriculture industry. This study investigates the use of convolutional neural networks (CNNs) and the YOLOv8 algorithm to enhance maize quality monitoring. Traditional maize sorting techniques are labor-intensive, time-consuming, and prone to mistakes. The incorporation of AI technology provides a solution to these difficulties. The study's goal is to create an automated system for recognizing and categorizing pest and disease-related maize quality concerns using machine learning and image recognition techniques. A CNN model was created utilizing labeled information to properly identify and categorize maize quality concerns into three groups. The model's performance was assessed using the YOLOv8 method, which is noted for its quick and accurate object identification capabilities. The training was done in the Google Colab environment, with pre-trained weights to speed up model convergence. The findings indicated significant increases in detection accuracy and efficiency. The model's overall accuracy was 92.4%, with class-specific accuracies of 88% for "Healthy," 65.5% for "Water Rot," and 100% for "Bug." The average Precision (mAP) was 92.7%, with an Intersection over Union (IoU) of 52.3%. Visual and statistical studies, such as F1-Confidence and RecallConfidence curves, offered information about the model's performance at different confidence levels. The findings emphasize the potential for AI-powered maize quality monitoring systems to improve agricultural practices, lower labor costs, and assure consistent and accurate quality evaluation. This study demonstrates the viability of implementing advanced deep-learning algorithms in real-world agricultural settings, opening the door for future crop monitoring and quality management advances.

Actions (login required)

View Item