Вы находитесь на странице: 1из 27

AUREAS: a tool for

recognition of anuran
vocalizations

1
AUREAS: a tool for recognition of
anuran vocalizations
William Gómez
PhD. Student
Universidad de Antioquia, Medellín, Colombia

Claudia Isaza Juan Manuel Daza


Professor - Universidad de Professor – Universidad de
Antioquia Antioquia

Sergio Gómez Carol Bedoya


Undergraduated Student (UdeA) PhD. Student – Cantembury
University

2
AGENDA

• Introduction
• AUREAS methodology.
• Study cases.

3
BIODIVERSITY

Ecosystem’s Biodiversity
Biodiversity
structure studies

Satellite images Passive


[Bawa, et. al., 2002] monitoring

Visual monitoring
[Pulido, et. al., 2017]
Traditional
monitoring
Bioacoustics
[Sueur, et. al., 2014]

4
BIODIVERSITY

Species
Biodiversity
Biodiversity Wild life
studies
Forests

Satellite images Passive


[Bawa, et. al., 2002] monitoring

Visual monitoring
[Pulido, et. al., 2017]
Traditional
monitoring
Bioacoustics
[Sueur, et. al., 2014]

5
BIOACOUSTICS
• No capture, manipulation
or contact.
• Large periods analyses.
• Low detectability.
• Cheaper.

• Same frequency calls.


• Within-Group variance.
• Enviromental noise

6
AUREAS Methodology

Recordings Segmentation Feature extraction

5
2.7
10

15
0.1
20 0.3
25

20 40 60 80 100 120
0.6
1.3
Species
Classification Feature selection
identification
2.7
25

20
0.1
15 0.3
Z

0.6
10 30

5 20

0
0 5 10 15 20 25
0
10

Y
1.3
X

7
AUREAS Methodology

Recordings Segmentation Feature extraction

5
2.7
10

15
0.1
20 0.3
25

20 40 60 80 100 120
0.6
1.3
Species
Classification Feature selection
identification
2.7
25

20
0.1
15 0.3
Z

0.6
10 30

5 20

0
0 5 10 15 20 25
0
10

Y
1.3
X

8
RECORDINGS

Nov – Dic 2012

1-min recordings using 10-min intervals.

1 1 1 1 1

9
AUREAS Methodology

Recordings Segmentation Feature extraction

5
2.7
10

15
0.1
20 0.3
25

20 40 60 80 100 120
0.6
1.3
Species
Classification Feature selection
identification
2.7
25

20
0.1
15 0.3
Z

0.6
10 30

5 20

0
0 5 10 15 20 25
0
10

Y
1.3
X

10
SEGMENTATION (1)
4
x 10

2 𝑓2 𝑛 = ෍ 𝑠[𝑛, 𝑤]
𝑤

1.5
5

𝑓1 [𝑤] = max (𝑠[𝑛, 𝑤])


n = [0,N] 4

1
80

3
70
60

2
50
40

0.5
30

1
20
10
0
0.2

0.18

0.16

0.14

0.12

0.1

0.08

0.06

0.04

0.02

0
5 10 15 20 25 30 35 40 45 50 55 -1
0 0.5 1 1.5 2 2.5
4
x 10

11
SEGMENTATION (2)
0.14 1

Bands selection
0.9
0.12
0.8

0.1
0.7

0.6
0.08

0.5
0.06
0.4

0.04 0.3

0.2
0.02
0.1

0 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80

Timestamps 4 5

selection 3
10

2
15

1
20

0
25

-1
0 0.5 1 1.5 2 2.5
4 20 40 60 80 100 120
x 10

12
AUREAS Methodology

Recordings Segmentation Feature extraction

5
2.7
10

15
0.1
20 0.3
25

20 40 60 80 100 120
0.6
1.3
Species
Classification Feature selection
identification
2.7
25

20
0.1
15 0.3
Z

0.6
10 30

5 20

0
0 5 10 15 20 25
0
10

Y
1.3
X

13
FEATURE EXTRACTION (FCC)

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8
0

2
0
2
5

4
6
10

8
10
15

12
20

14
16
25

18
20
20 40 60 80 100 120

𝑁
2 𝜋
𝑦(𝑘) = ෍ 𝑥(𝑛)cos (2𝑛 − 1)(𝑘 − 1)
𝑀 2𝑁
𝑛=1

14
AUREAS Methodology

Recordings Segmentation Feature extraction

5
2.7
10

15
0.1
20 0.3
25

20 40 60 80 100 120
0.6
1.3
Species
Classification Feature selection
identification
2.7
25

20
0.1
15 0.3
Z

0.6
10 30

5 20

0
0 5 10 15 20 25
0
10

Y
1.3
X

15
CLASIFICATION (LAMDA)

[Bedoya, 2014]

16
AUREAS Methodology

Recordings Segmentation Feature extraction

5
2.7
10

15
0.1
20 0.3
25

20 40 60 80 100 120
0.6
1.3
Species
Classification Feature selection
identification
2.7
25

20
0.1
15 0.3
Z

0.6
10 30

5 20

0
0 5 10 15 20 25
0
10

Y
1.3
X

17
SPECIES IDENTIFICATION

18
SPECIES IDENTIFICATION

Activity patterns

19
VIDEO

20
ANURAN RECOGNITION (1)
Class Samples Species

Subocularis 70 Dendropsophus
subocularis
Sachatamia 74 Sachatamia
Prosoblepon 66 Espadarana prosoblepon
Norandinus (1) 66 Dendropsophus
Norandinus (2) 89 Norandinus
Diasporus 112 Diasporus
Colostethus 289 Colostethus
Empty calls 2000 Noise

21
AVIAN RECOGNITION (2)

FEATURE EXTRACTION

• Wavelets
• LPC (Linear predictive codes)
• BFCC (Bark Frequency Cepstral
Coefficients)
• Mel Frequency Cepstral Coefficients
(MFCC)
• Perceptual Linear Prediction (PLP)

22
ANURAN RECOGNITION (1)

RNA LAMDA

Features Acc. Val Acc. Test F1 score Features Acc. Val Acc. Test F1 score
Wavelets 0,88 ± 0,01 0,89 ± 0,02 0,72 ± 0,02 PLP 0,65±0,01 0,52±0,04 0,59±0,04
PLP 0,93 ± 0,04 0,93 ± 0,04 0,86 ± 0,04 LPCC 0,60±0,03 0,54±0,04 0,59±0,06
LPCC 0,95 ± 0,01 0,95 ± 0,02 0,86 ± 0,05 OUR MFCC 0,63±0,01 0,58±0,01 0,59±0,06
MFCC 0,92 ± 0,02 0,90 ± 0,03 0,85 ± 0,01
BFCC 0,93 ± 0,02 0,91 ± 0,01 0,85 ± 0,02
OUR MFCC 0,95 ± 0,01 0,94 ± 0,04 0,88 ± 0,03

23
AVIAN RECOGNITION (1)

24
AVIAN RECOGNITION (2)

FEATURE EXTRACTION FEATURE SELECTION

• Wavelets • PCA
• LPC (Linear predictive codes) • Laplace score
• BFCC (Bark Frequency Cepstral • Spectral feature selection
Coefficients)
• Mel Frequency Cepstral Coefficients
(MFCC)
• Perceptual Linear Prediction (PLP)

25
AVIAN RECOGNITION (2)

RNA LAMDA

26
Conclusions and future work

Species identification is achievable using bioacoustics


and classification methods.

Selection of variables by filtering? Or using supervised


on a group of data (performance).

It is necessary to analyze the option of evolving the


methods with a previous supervised classification and
then create new classes for the case of birds.

27

Вам также может понравиться