Comparative assessment of feature selection and classification techniques for visual inspection of pot plant seedlings

Author(s)

Silva, L. O. L. A. ^[1] ; Koga, M. L. ^[1] ; Cugnasca, C. E. ^[1] ; Costa, A. H. R. ^[1]

Affiliation

^[1] Univ Sao Paulo, Agr Automat Lab LAA, BR-05508900 Sao Paulo - Brazil

• Document type: Journal article
• Source: COMPUTERS AND ELECTRONICS IN AGRICULTURE; v. 97, p. 47-55, SEP 2013.
• Web of Science Citations: 14

Abstract

Homogeneity plays an important role in ornamental plant and flower production. As assessing the quality of seedlings is an effective way of predicting plant growth performance, a vision system capable of performing this task is desirable. Yet, the optical sorting of agricultural products must find ways to incorporate knowledge from human experts into the computational solution. Our aim is evaluating feature selection techniques with respect to the performance of vision-based inspection and classification of pot plant seedlings. A large feature set was initially obtained from seedlings images and several subsets were generated with various features selection techniques. The performance of each subset was compared to some of the most popular classifiers in the literature: Naive Bayes, k-Nearest Neighbors, Logistic Regression, C4.5, Random Forest, Multilayer Perceptron as well as Partial Least Squares and Support Vector Machine Discriminant Analysis. The best classifier and subset configuration is presented; our results show that feature selection was indeed advantageous, generating accuracy gains of up to 7.4%. (c) 2013 Elsevier B.V. All rights reserved. (AU)

• FAPESP's process: 12/02190-9 - Knowledge Transfer among Tasks in Reinforcement Learning
• Grantee: Marcelo Li Koga
• Support Opportunities: Scholarships in Brazil - Master
• FAPESP's process: 11/19280-8 - CogBot: integrating perceptual information and semantic knowledge in cognitive robotics
• Grantee: Anna Helena Reali Costa
• Support Opportunities: Regular Research Grants