Innovazioni biofotoniche in agricoltura: analisi iperspettrale e della forma per immagini
Author(s)
Menesatti, Paolo
Date Issued
December 20, 2010
Type
Doctoral Thesis
Abstract
Biophotonics includes all the technological and engineering disciplines that use the
electromagnetic radiation as the main carrier of information (photonics - 50 billion €,
246.000 workers in the EU in 2005) in earth sciences and biosystems applications.
The use of biophotonics in agricultural engineering discipline is growing in relation to
the high innovation rate (based on high-tech sectors: optics, electronics, computers
and algorithms) and the needs of research, production monitoring and control
systems, highly informative, multiparametric, rapid, non destructive, for applications
in the field (on-the-go) or in the process line (on-line). Numerous studies have been
analyzed and reported for agricultural applications in: robotics (vision), tractors and
self-propelled machines (automatic or assisted guide), agricultural field machines
(fertilizers and pesticide precise distribution), test and certification of operational
characteristics and work quality, post-harvest machines (fruit selection based on
external and internal qualities), quality of agricultural products (meat, fruit and
vegetables, fish) and food (cheese, bread).
Among such a variety of applications this work consider two of the most innovative
technical-methodological sector of application, shape imaging and hyperspectral
imaging, which have been reported in experimental applications published or
submitted for publication (mainly on international ISI journals).
A specific review (Shape Analysis of Agricultural products by computer vision - a
review of recent research advances) explored the technical-methodological
complexity and the scientific applications of the shape analysis of agricultural
products. This technique revealed to be useful for the objective selection of samples
variety quality (for example, distinguishing the geographical origin of fruits), the
determination of thresholds for damage and defects of shape and form and as a
support to genetists for clonal and varietal comparison. Experimentally, the technique
has been applied by combining the sample profiles, extracted from the images of
agricultural products, with the supervised multivariate classificatory model analysis
(Partial Least Square Discriminant Analysis - PLSDA) applied to the harmonic
coefficients of the equations that mathematically describe the profile, calculated
through the algorithm EFA (Elliptical Fourier analysis). This system was the first
3
time applied by the authors in the food industry. It has proved very powerful and
discriminating, allowing the correct classification of almond fruit (kernel and in-shell)
with an average value of 90% between groups of similar shape on a total of 18
cultivars (Quantitative method for shape description of almond cultivars [Prunus
amygdalus Batsch]). The same analysis (Shape-based methodology for multivariate
Discrimination Among Italian hazelnut cultivars) was applied on four traditional
hazelnut cultivars (round: Tonda di Giffoni, Tonda Romana, oblong: San Giovanni,
Mortarella). The classification results with over 95% between the cultivar within each
group. Moreover, 17 different genotypes of Tarocco oranges (Tarocco Discrimination
of sweet orange [Citrus sinensis (L.) Osbeck] varieties using Elliptic Fourier based
opto-electronic analysis of fruit shape) were analyzed with this technique (EFA +
PLSDA), to classify 5 morphological groups in a more objective way than the
reference selaction technique which sorts them manually (IPGRI or Citrus Industry).
Finally, this novel technique has been applied, with a good success (average 69%), to
assess the degree of attack of two phungi micotoxins responsible for Fusarium head
blight of wheat by measuring variations in the kernels shape (Application of
morphometric image analysis system to evaluate the incidence of fusarium head
blight infected wheat kernels).
Hyperspectral imaging is one of the most powerful techniques to resolve in spectral
dimension all the pixels of an image. The subject of analysis is characterized by
spectral fingerprint and spatial properties (morphological, structural). Apart from
remote sensing, the applications range from environmental monitoring (proximal
sensing), to field machinery or to the agricultural products and food quality analysis
and selection. An innovative determination of the maturity degree of Golden apples
(Supervised multivariate analysis of hyperspectral NIR images to evaluate the starch
index of apples) is the first experimental work reported. It was based on the
determination of the traditional starch index, calculated innovatively through a
hyperspectral near infrared (NIR) imaging. The obtained good results indicate the
possibility to replace the reference test, involving the use of chemicals (iodineiodide),
with a rapid and non-contact technique. The second activity (Quality
evaluation of fish by hyperspectral imaging) reports a large overview of hyperspectral
imaging application for fish quality analysis. The experimental part aimed to estimate
4
fish freshness of farmed fish in terms of days of chilled conservation. The article
reports the first literature example of the combination of shape and hyperspectral
imaging in agricultural and food systems. In this way, it was possible to extract the
more informative spectral bands and surface area of the fish, for freshness estimation.
The fast non-contacting system was able to detect the fish freshness, correctly
classifying (79.4%) fish stored only 1-2 days (fresh product) from fish stored 3-5 days
(fish not fresh, but still perfectly edible).
Additional information
Dottorato di ricerca in Meccanica agraria
File(s)![Thumbnail Image]()
Name
pmenesatti_tesid.pdf
Size
12.81 MB
Format
Adobe PDF
Checksum (MD5)
7554f8542932b7ceec1fa44f3cd8049c