Digital Microscopic Imaging for Citrus Peel Disease Classification Using Color Texture Features

Abstract

Listen

Translate article

Citrus industry is a major part of Florida's agricultural and natural resource economy. Proper and timely identification and control of citrus disease could assure fruit quality and safety, improve production, and minimize economic losses. This study was aimed to develop a machine vision based method for detecting common citrus peel diseases using color texture feature analysis of microscopic images. A digital microscope system was applied to acquire RGB images at magnifications 5x, 20x, and 50x from grapefruits with six peel conditions (i.e., normal, canker, scab, greasy spot, melanose, and insect damage). Thirty samples for each peel condition were selected. For each fruit sample, 39 texture features were computed from the transformed hue, saturation, and intensity region of interest images by using color co-occurrence method. Algorithms based on a stepwise discriminant analysis were performed to select useful texture features. Three reduced feature models include hue, saturation, and intensity (H, S, I), hue and saturation (H, S), and intensity (I) only. Significant eliminations of redundant texture features were accomplished through the stepwise discriminant analysis. Reduced features model HSI and a model using all 39 HSI texture features outperformed model HS and I only. Best overall classification accuracy (95.0%) was achieved for selected HSI model and all HSI model for magnification 20x dataset. A stability test for the classification model with the best performance was accomplished by ten run using randomly selected training and testing samples. The average and standard deviation of the classification accuracy were 93.3% and 3.2%, respectively. The results suggested that a reduced hue, saturation and intensity texture features model coupled with microscope imaging with magnification 20x was a good tool to differentiate citrus peel conditions.