Možnost uporabe metod strojnega učenja pri napovedovanju razvoja krovne barve jabolk (malus domestica borkh.)

Germšek, Blaž

| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

First page > Show document

Show document

Title:	Možnost uporabe metod strojnega učenja pri napovedovanju razvoja krovne barve jabolk (malus domestica borkh.)
Authors:	ID Germšek, Blaž (Author) ID Rozman, Črtomir (Mentor) More about this mentor... ID Unuk, Tatjana (Comentor)
Files:	DOK_Germsek_Blaz_2017.pdf (3,56 MB) MD5: E58D67FD32558E4FBE0B124AC591828E
Language:	Slovenian
Work type:	Doctoral dissertation
Organization:	FKBV - Faculty of Agriculture and Life Sciences
Abstract:	Napovedovanje razvoja parametrov kakovosti je v kmetijski proizvodnji ob zahtevah sodobnega potrošnika postalo zelo pomembno. Tipičen parameter, ki ga potrošnik neposredno povezuje s kakovostjo, je barva kožice plodov. V raziskavi smo se osredotočili na možnosti napovedovanja razvoja in intenziviranja krovne barve jabolk, katere intenzivnost smo določili z barvnim parametrom a. Za metodo strojnega učenja smo uporabili nadzorovano učenje, modele pa generirali s šestimi različnimi odločitvenimi drevesi (Decision Stump, J48, LMT, Random Forest, Random Tree in Rep Tree). Namen raziskave je bil zgraditi modele, ki bodo na podlagi vremenske napovedi omogočali sprejemljivo natančnost (< 60-% natančnost) napovedi razvoja krovne barve oz. barvnega parametra a za tri obravnavane sorte jabolk (‘Gala Brookfield’, ‘Fuji Kiku 8’ in ‘Braeburn Mariri red’). Pri sorti ‘Gala Brookfield’ smo najbolj natančne modele generirali z uporabo odločitvenega drevesa J48 (89,13-% natančnost). Pri sorti ‘Fuji Kiku 8’ smo najbolj natančen model dobili z uporabo odločitvenega drevesa LMT (91,73-% natančnost), s tem modelom (LMT) smo dobili najbolj natančen model napovedi barvnega parametra a* tudi za sorto ‘Braeburn Mariri red’ (96,65-% natančnost). Model LMT je izmed vseh generiranih modelov (36) in obravnavanih sort (3) dal najbolj natančno napoved barvnega parametra a*, kar pripisujemo predvsem najbolj natančnim podatkom, ki smo jih s statistično obdelavo dobili pri sorti ‘Braeburn Mariri red’. Ugotovili smo, da je uporabnost napovedovalnih modelov odvisna predvsem od natančnosti vremenske napovedi. Pri 7-dnevni uradni vremenski napovedi, ki je bila uporabljena pri naših ekspertnih modelih, se natančnost modelov v povprečju zniža za 10,73 %. Znižanje natančnosti napovedovalnih modelov pripisujemo zelo variabilni vremenski napovedi, s katero smo v času raziskave razpolagali.
Keywords:	strojno učenje, napovedovanje, kakovostni parametri, jabolka, odločitvena drevesa
Place of publishing:	Maribor
Year of publishing:	2017
PID:	20.500.12556/DKUM-64866
NUK URN:	URN:SI:UM:DK:N3RA7T08
Publication date in DKUM:	21.02.2017
Views:	3741
Downloads:	334
Metadata:
Categories:	FKBV
:	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:	Using machine learning methods for apple skin color prediction (Malus domestica Borkh.)
Abstract:	Prediction of the development of quality parameters in agricultural production with the requirements of the modern consumer has become very important. A typical parameter which the consumer directly relates to the quality is the colour of the fruit skin. In this thesis, we focused on the possibility of predicting the development and intensification of colour of apples, whose intensity was determined by the colour parameter a . We used supervised learning as a machine learning method. Models generated six different decision trees (Decision Stump, J48, LMT, Random Forest, Random Tree and Tree Rep). The purpose of the research was to build models that will base on weather forecasts which allow for acceptable accuracy (<60% accurate) of predictions on the development of the fruit colour or colour parameter a of the three apple varieties ('Brookfield Gala', 'Fuji Kiku 8' and 'Braeburn Mariri red'). Regarding the variety 'Gala Brookfield', we have generated the most accurate models using a decision tree J48 (89.13-% accuracy). In addition, the variety 'Fuji Kiku 8' obtained the most accurate model by using a decision tree LMT (91.73-% accuracy). While using this model (LMT), we get the most accurate prediction model of colour parameter a * for the variety 'Braeburn Mariri red' (96.65-% accuracy). Model LMT, out of all generated models (36) and the varieties (3) considered, predicted colour parameters a * the most accurately. That can be ascribed to the most accurate data, statistically processed with variety 'Braeburn Mariri red'. We can conclude that the applicability of the predictive models depends mainly on the accuracy of weather forecasts. Using a 7-day official weather forecast with our expert models, shows that the accuracy of the models is reduced by 10.73% on average. A reduction in accuracy of predictive models is attributed to the very variable weather forecast, which was at disposal during our research.
Keywords:	machine learning, prediction, quality parameters, apple, decision trees

Comments

Leave comment

You must log in to leave a comment.

Comments (0)

0 - 0 / 0

There are no comments!

Back