AI-Assisted Cotton Grading: Active and Semi-Supervised Learning to Reduce the Image-Labelling Burden.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate crop quality assessment during harvesting is crucial for determining market value and processing needs.
• Machine learning (ML) models offer potential for crop grading, but high data labeling costs hinder their deployment.
• Current ML applications in agriculture face challenges with extensive manual data annotation.

Purpose of the Study:

• To evaluate the effectiveness of semi-supervised and active learning in reducing labeling effort for Egyptian cotton grading.
• To compare the classification accuracy and data requirements of active learning against supervised and semi-supervised learning.
• To assess the impact of active learning on the time efficiency of ML model development for crop grading.

Main Methods:

• Development of Random Forest classification models using supervised, semi-supervised, and active learning strategies.
• Application of these models for grading Egyptian cotton lint samples.
• Quantitative analysis of classification accuracy, labeled data volume reduction, and time efficiency.

Main Results:

• Active learning models achieved higher accuracy (82.85-85.33%) compared to supervised (80.20-82.66%) and semi-supervised learning (81.39-85.26%).
• Active learning reduced the required labeled data volume by up to 46.4%.
• Labeling time for cotton lint samples was decreased from 422.5 to 177.5 minutes using active learning.

Conclusions:

• Active learning is a highly effective strategy for minimizing data labeling effort in crop grading.
• This approach enables the development of accurate and time-efficient ML models for agricultural quality assessment.
• Active learning presents a promising solution to overcome the data annotation bottleneck in applying ML to food and industrial crops.