Aplikacija za avtomatizirano prepoznavanje listov dreves in grmov kritosemenk za operacijski sistem Android

Puhmeister, Sandro

| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

First page > Show document

Show document

Title:	Aplikacija za avtomatizirano prepoznavanje listov dreves in grmov kritosemenk za operacijski sistem Android
Authors:	ID Puhmeister, Sandro (Author) ID Pesek, Igor (Mentor) More about this mentor... ID Potočnik, Božidar (Comentor)
Files:	UN_Puhmeister_Sandro_2016.pdf (2,77 MB) MD5: C14E0FE3676A2252A061C075046BF6A6
Language:	Slovenian
Work type:	Undergraduate thesis
Typology:	2.11 - Undergraduate Thesis
Organization:	FNM - Faculty of Natural Sciences and Mathematics
Abstract:	V tem diplomskem delu smo implementirali aplikacijo za avtomatizirano prepoznavanje listov dreves in grmov kritosemenk za operacijski sistem Android, pri čemer smo postopek prepoznavanja vrste razdelili na štiri ključne faze: zajem slik, obdelavo slik, luščenje značilnic in določitev vrste. Pri zajetju slik smo uporabili privzeto aplikacijo operacijskega sistema Android, kjer uporabnik zajame dve fotografiji (fotografira spodnjo in vrhnjo površino lista) za obdelavo slik. Pri obdelavi slik smo razvili lasten algoritem za segmentacijo regije interesa v RGB barvnem prostoru. Vsako segmentirano regijo interesa smo nato opisali z Gaborjevimi značilnicami tekstur. Gaborjev deskriptor za luščenje Gaborjevih značilnic smo uporabili iz odprtokodne Java knjižnice LIRE 1.0b2 (Lux Mathias, 2008). Vektorje značilnic smo nato združili in jih pretvorili v instance za prepoznavanje vrste na stroju podpornih vektorjev (angl. Support vector machine ali SVM). Za učenje SVM klasifikatorja smo uporabili brezplačen paket Weka 3.8 (Alexis Joly, 2015). Učili smo večrazredni SVM klasifikator s sekvenčnim minimalnim optimizacijskim algoritmom (angl. Sequential minimal optimization algorithm ali SMO) s polinomsko jedrno funkcijo (poglavje 6.3). Klasifikator smo kalibrirali s funkcijo K zvezd (K*). Najprej smo klasifikator testirali s križno validacijo na učnih vzorcih ter dobili natančnost nad 90%, nato pa smo klasifikator testirali še dvakrat na 100 neznanih vzorcih, pri čemer bi morali biti vsi neznani vzorci klasifikatorju znani. Natančnost klasifikatorja je bila v drugem in tretjem testiranju še zmeraj visoka saj je znašala nad 80%. Natančnosti klasifikatorja na različnih mobilnih napravah z operacijskim sistemom Android nismo uspeli testirati zaradi prevelikega števila Android naprav na trgu, kjer ima vsaka naprava lastne nastavitve. Prvi testi na napravi Samsung Galaxy Tab S 8.4 so pokazali, da je avtomatizirano določevanje vrste lahko uspešno, če so fotografije visoke kakovosti in so zajete v ugodnih svetlobnih pogojih (na dnevni svetlobi) ter manj uspešno, če ti pogoji niso izpolnjeni.
Keywords:	Gaborjeve značilnice, računalniški vid, luščenje značilnic, prepoznavanje vrst dreves in grmov, avtomatizirano prepoznavanje vrst.
Place of publishing:	Maribor
Publisher:	[S. Puhmeister]
Year of publishing:	2016
PID:	20.500.12556/DKUM-62273
UDC:	004.9:582.091(043.2)
COBISS.SI-ID:	22581512
NUK URN:	URN:SI:UM:DK:1IAT2UP3
Publication date in DKUM:	27.09.2016
Views:	2854
Downloads:	182
Metadata:
Categories:	FNM
:	Copy citation

Average score:	(0 votes)
Your score:	Voting is allowed only for logged in users.
Share:

Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Secondary language

Language:	English
Title:	Application for automated identification of leaves of angiosperm trees and shrubs on Android operating system
Abstract:	In this thesis, an application for automated recognition of leaves of angiosperm trees and shrubs for Android operating system was implemented, whereby species identification process was divided into four key stages: image capturing, image processing, feature extraction and identification of species. A preinstalled Android application was used for image capturing, which enables the user to capture two images (one of the top and one of the bottom of a leaf) required for feature extraction. For image processing, a specialized algorithm for image segmentation in RGB color space was developed. Each segmented region of interest was later described with Gabor texture features. Gabor descriptor for obtaining Gabor features was used from LIRE 1.0b2 (Lux Mathias, 2008), an open source Java library. The feature vectors were then concatenated and converted into instances for a support vector machine (SVM) model. The SVM model was trained for multiclass classification using Weka 3.8 package (Alexis Joly, 2015) with a sequential minimal optimization algorithm (SMO) function and a polynomial kernel function. SVM classifier was calibrated with the K-star (K*) function. First, the classifier was tested using 10-fold cross validation on training samples, where classifier accuracy was slightly above 90%, afterwards, the classifier was tested two more times on 100 unknown samples, whereby all of the unknown samples should be recognized by the classifier. The accuracy of the classifier was still high, above 80%. The classifier could not be tested on Android mobile devices due to a large number of Android devices on the market and each of them having its own unique hardware settings. First tests on a Samsung Galaxy Tab S 8.4 device showed that automated recognition of species should have high success rates if the images are of good quality and taken in good lightning conditions (in daylight), and low success rates if those conditions are not fulfilled.
Keywords:	Gabor features, computer vision, feature extraction, recognition of trees and shrubs, automated recognition of species.

Comments

Leave comment

You must log in to leave a comment.

Comments (0)

0 - 0 / 0

There are no comments!

Back