On Combining Boosting with Rule-Induction for Automated Fruit Grading

摘要: The automation of post-harvest fruit grading in the industry is a problem that receiving considerable attention realm computer vision and machine learning. Classification accuracy with automated systems this domain challenge given inherent variability visual appearance its quality-determining features. While paramount importance, usability interpretability learning solutions to operators are also crucial since many sophisticated algorithms involve numerous tunable parameters often “black-boxes”. This research presents generalizable solution balances need for high by decomposing into sub-tasks. A powerful boosting algorithm (AdaBoost.ECC) low employed fruit-surface characteristics. classification outputs then become inputs rule-induction (RIPPER FURIA), generating human-interpretable rule sets amenable review revisions operators. Using seven datasets different varieties, performance proposed method was compared against manually calibrated commercial fruit-grading system. results showed system able match machines experts having years experience, while providing simpler possessing yielding knowledge discovery.