Agriculture Reference
In-Depth Information
(a)
(b)
FIGURE 7.44
(a) Detection of green fruits; (b) detection of mature fruits.
methods (8 points algorithm and Plucker coordinate system) were used to recon-
struct the 3-D citrus canopy. The Plucker coordinates system showed a better result
than 8 points algorithm.
7.6.5 M
ACHINE
V
ISION
U
SED
IN
D
ETECTING
C
ITRUS
G
REENING
ON
L
EAVES
Pydipati et al. (2006) developed a disease detection approach for citrus leaves using
color vision-based texture analysis. The image data of infected leaves from various
diseases common in citrus were collected, and texture features were extracted using
the color co-occurrence method reported by Burks et al. (2000). Significant texture
features were selected and used to train both statistical classifiers and neural network
classifiers. Classification accuracies approaching 97% were achieved under labora-
tory conditions. Kim (2011) developed a machine vision-based method for detecting
citrus greening on leaves using color texture features under controlled lighting in
order to discriminate between citrus greening and leaf nutrient deficiency conditions
commonly confused with citrus greening. This approach used low-level magnifica-
tion to enhance features and was conducted in a laboratory setting achieving clas-
sification accuracies above 95%.
Mishra et al. (2007) developed a spectral method for the detection of Citrus
Huanglongbing (HLB). Canopy reflectance spectral data were collected with an
ASD FieldSpec
®
spectroradiometer. Using techniques of spectral wavelength dis-
criminability, spectral derivative analysis, and spectral ratio analysis, the research
aimed at identifying optimal wavebands for accurate detection of HLB in citrus. It
was concluded that the visible region (400-700 nm) has good (0.89-0.85) discrimi-
nation. Results from the finite difference second derivative method revealed that
wavelengths of 480, 590, 754, 1041, and 2071 nm have the potential to differentiate
HLB. Using the spectral ratio analysis, it was found that the reflectance of HLB-
infected trees at 530-564 nm was higher than that of healthy trees. A second sensitive
point was observed at 710-715 nm. Sankaran and Ehsani (2010) used mid-infrared
(MIR) spectroscopy to analyze and detect HLB-infected citrus leaves. The healthy,
nutrient-deficient, and HLB-infected leaves were ground under liquid nitrogen and
analyzed using a portable MIR instrument. The preprocessed data were analyzed
with principal component analysis and the samples were classified using quadratic
