26 June 2017 Feature selection from hyperspectral imaging for guava fruit defects detection
Author Affiliations +
Proceedings Volume 10335, Digital Optical Technologies 2017; 103351E (2017) https://doi.org/10.1117/12.2270137
Event: SPIE Digital Optical Technologies, 2017, Munich, Germany
Abstract
Development of technology makes hyperspectral imaging commonly used for defect detection. In this research, a hyperspectral imaging system was setup in lab to target for guava fruits defect detection. Guava fruit was selected as the object as to our knowledge, there is fewer attempts were made for guava defect detection based on hyperspectral imaging. The common fluorescent light source was used to represent the uncontrolled lighting condition in lab and analysis was carried out in a specific wavelength range due to inefficiency of this particular light source. Based on the data, the reflectance intensity of this specific setup could be categorized in two groups. Sequential feature selection with linear discriminant (LD) and quadratic discriminant (QD) function were used to select features that could potentially be used in defects detection. Besides the ordinary training method, training dataset in discriminant was separated in two to cater for the uncontrolled lighting condition. These two parts were separated based on the brighter and dimmer area. Four evaluation matrixes were evaluated which are LD with common training method, QD with common training method, LD with two part training method and QD with two part training method. These evaluation matrixes were evaluated using F1-score with total 48 defected areas. Experiment shown that F1-score of linear discriminant with the compensated method hitting 0.8 score, which is the highest score among all.
© (2017) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Mohd. Zubir Mat Jafri, Sou Ching Tan, "Feature selection from hyperspectral imaging for guava fruit defects detection", Proc. SPIE 10335, Digital Optical Technologies 2017, 103351E (26 June 2017); doi: 10.1117/12.2270137; https://doi.org/10.1117/12.2270137
PROCEEDINGS
11 PAGES


SHARE
Back to Top