Automated system for tobacco quality classification and pricing using image processing and machine learning algorithms: A Case of Tobacco Auction Floor.

Author
Ndlovu, Dereck
Title
Automated system for tobacco quality classification and pricing using image processing and machine learning algorithms: A Case of Tobacco Auction Floor.
Abstract
Automated Tobacco Leaf Grading and Pricing System to Increase Auction Floor Equity in Zimbabwe. A vital component of Zimbabwe's agrarian economy, the tobacco business sometimes struggles with subjective pricing and manual leaf sorting, which results in inefficiencies and discontent among farmers. By automating quality evaluation and pricing determination, this research improves objectivity and transparency by creating a system that combines image processing and machine learning. Pre-processing (blob detection, threshold, morphological operations), feature extraction (colour histograms, texture analysis), and a machine learning core utilizing Convolutional Neural Networks (CNNs) are the main modules that the system uses to handle uploaded leaf photos. According to empirical validation, the CNN model outperforms FNN (97%), Decision Trees (97%), and DenseNet121 (75%), and it outperforms SVR (75.3%) and Random Forest (39.75%) in price regression with an accuracy of 76.38%, surpassing SVR (75.3%) and Random Forest (39.75%). While white-box testing confirmed algorithmic soundness, extensive black-box testing confirmed end-to-end functioning, including image upload, grading, price, and voucher generation. The system's usability and output clarity were highlighted in a pilot study with 11 farmers, where it received a user satisfaction rating of 4.56/5. This study shows that using CNNs to automate pricing and grading greatly lowers human bias, speeds up auction procedures, and increases farmer trust. Large-scale pilot deployment, hybrid model refinement (e.g., WOA-Stacking ensembles), and ethical supervision frameworks are among the recommendations. The system provides Zimbabwe and similar agro-economies with a scalable model for fair tobacco trade.
Date
June 2025
Publisher
BUSE
Keywords
Tobacco Quality
Image Processing
Supervisor
Mr. D. Ndumiyana
Item sets
Department of Computer Science
Media
Ndlovu, Dereck.pdf

Part of Automated system for tobacco quality classification and pricing using image processing and machine learning algorithms: A Case of Tobacco Auction Floor.