Transcript METHODOLOGY

Artificial Vision for the Recognition
of Exportable Mangoes
by Using Neural Networks
Hugo Froilán Vega Huerta
UNMSM
Ana María Huayna Dueñas
Antecedents
Antecedents
3
Antecedents
Percentages of Export for Types of Mangoes
4
Antecedents
5
Antecedents
BRIOFRUIT staff are separating mangoes that won’t be exportable.
6
Antecedents
THE PROBLEM
Statistics plant selection (purification of mangoes malformed)
8
THE PROBLEM
¿ How the neural networks allow the
recognition of the quality of export mangoes
in Biofruit?
9
OBJECTIVES
•Achieve to train a neural network that is able of
recognize export mangoes.
•Achieve to reduce the margin of error from 6.5% to
3%.
10
Theoretical Framework
DEFINITION [James A. Anderson 2007]
Is a set of units of processing called Neurons, cells or nodes, interconnected to each other
by bonds of communication direct called connections, with the purpose of receiving input
signals, process them and emit output signals. Each connection is associated to a weight
that they represent the knowledge of the RN
They are models Mathematical inspired in the operation of the biological neural networks,
consequently, central processing units of a RNA, will be the Artificial Neurons.
Next we present the graphic representation of a RNA
11
Theoretical Framework
RN TRAINING [Edgar N. Sánchez, 2006]
It consists on presenting to the system a set of pairs of data, representing the input and
the wanted output for this input. This set receives the name of group of training. The
objective is to try to minimize the error between the Wanted output and the current one.
The weights are adjusted in function of the difference between the wanted values and the
obtained output values.
12
STATE OF THE ART
•Doctoral Thesis - Facial recognition techniques using neural
networks (Enrique Cabello P. – Politécnica de Madrid University,
2004)
•Master Thesis - Techniques to improve voice recognition in the
Presence of Out of Vocabulary Speech (Heriberto Cuayáhuitl Portilla Las Américas de Puebla University Foundation)
•Article- Shape Recognition of Film Sequence with Application of
Sobel Filter and Backpropagation Neural Network (A. Glowacz and
W. Glowacz 2008)
13
STATE OF THE ART
COMPARATIVE EVALUATION OF METHODS OF PATTERN
RECOGNITION (Eybi Gil Z, 2010)
14
STATE OF THE ART
COMPARATIVE EVALUATION OF METHODS OF PATTERN
RECOGNITION (Eybi Gil Z, 2010)
15
METHODOLOGY
Neural Network For
Recognition of Exportable Mangos
METHODOLOGY
Artificial Vision
17
METHODOLOGY
Artificial Vision
18
METHODOLOGY
Artificial Vision
19
METHODOLOGY
Artificial Vision
20
METHODOLOGY
Artificial Vision
21
METHODOLOGY
Recognition of Exportable Mangos
Functional dependency between input and output data
in a Neural Network
22
METHODOLOGY
Architecture of the NN for the recognition of mangoes
23
METHODOLOGY
Knowledge Base for Neuronal Network Training
24
METHODOLOGY
Knowledge Base for Neuronal Network Training
25
METHODOLOGY
Knowledge Base for Neuronal Network Training
26
METHODOLOGY
Neural Network Training
27
METHODOLOGY
Recognition of Exportable Mangoes
¡ Exportable Mangoes !
¿ Exportable Mangoes? ? ? ?
28
METHODOLOGY
Recognition of Exportable Mangoes
29
METHODOLOGY
Recognition of Exportable Mangoes
We execute the program of Recognition
Output information
Interpretation
30
Automated System
31
Automated System
32
Automated System
33
Automated System
34
CONCLUSIONS AND
RECOMMENDATIONS
C1: It is feasible to train neural networks for recognition of
exportable mangoes
C2: The recognition of exportable mangoes by Artificial Neural
Networks has reduced the margin of error of 6.7% 2.3%
R1: In processes where you need recognize one or more species,
types or subsets of elements where the elements that belong to
each type are different but have a common pattern that
identifies them, we recommend to use NN of Multilayer
Perceptron type with algorithm Backpropagation.
R2: For the success of the pattern recognition is recommended to
analyze and identify properly the characteristic of similarity
between units of the same pattern and the differences between
elements of other patterns
35
THANK YOU VERY MUCH
36