Difference between revisions of "ISSS608 2017-18 T3 Kiriti Yelamanchali Q2"
Kiritiy.2017 (talk | contribs) |
Kiritiy.2017 (talk | contribs) |
||
| (One intermediate revision by the same user not shown) | |||
| Line 24: | Line 24: | ||
| style="font-family:Century Gothic; font-size:100%; solid #1B338F; background:#00BFFF; text-align:center;" width="14.28%" | | | style="font-family:Century Gothic; font-size:100%; solid #1B338F; background:#00BFFF; text-align:center;" width="14.28%" | | ||
| − | |||
| − | |||
| − | |||
; | ; | ||
| − | [[ISSS608_2017- | + | [[ISSS608_2017-18_T3_Kiriti_Yelamanchali_Q3| <font color="#FFFFF">'''Conclusion'''</font>]] |
| style="font-family:Century Gothic; font-size:100%; solid #00BFFF; background:#00BFFF; text-align:center;" width="14.28%" | | | style="font-family:Century Gothic; font-size:100%; solid #00BFFF; background:#00BFFF; text-align:center;" width="14.28%" | | ||
| Line 57: | Line 54: | ||
4. The high quality sound files are selected, and split as train, test and validation datasets. | 4. The high quality sound files are selected, and split as train, test and validation datasets. | ||
| − | [[File:Kiriti ml 2.png | | + | [[File:Kiriti ml 2.png | 600 px]] |
Librosa is used to extract the features of each file and stored in dataframe. | Librosa is used to extract the features of each file and stored in dataframe. | ||
| − | [[File:Kiriti ml 3.png | | + | [[File:Kiriti ml 3.png | 600px]] |
5. Deep learning model is trained. The hyper parameters are tuned until a satisfactory level of accuracy on train and test data are obtained. | 5. Deep learning model is trained. The hyper parameters are tuned until a satisfactory level of accuracy on train and test data are obtained. | ||
| − | [[File:Kiriti dl.png | | + | [[File:Kiriti dl.png | 600px]] |
Hyper parameters tuning: | Hyper parameters tuning: | ||
| − | [[File:Kiriti ml 7.png | | + | [[File:Kiriti ml 7.png | 600 px]] |
Accuracy is observed for the model trained. | Accuracy is observed for the model trained. | ||
| − | [[File:Kirit acc.png | | + | [[File:Kirit acc.png | 600 px]] |
6. Once happy with the model, the birds in the given test data set are observed. | 6. Once happy with the model, the birds in the given test data set are observed. | ||
| − | [[File:Kiriti Test sightings.png | | + | [[File:Kiriti Test sightings.png | 600 px]] |
==Conclusion== | ==Conclusion== | ||
* It is observed that 4 out of the 15 test birds are Rose-Crested-Blue-Pipits. The rest are as follows: | * It is observed that 4 out of the 15 test birds are Rose-Crested-Blue-Pipits. The rest are as follows: | ||
| + | |||
| + | [[File:Kiriti Test sightings.png | 600 px]] | ||
Latest revision as of 00:07, 9 July 2018
VAST CHALLENGE 2018: MC1
|
|
|
|
|
|
|
Deep Learning Model to recognize bird calls
For this analysis, a deep learning model is built to recognize the bird calls.
- Python library librosa is ised for this purpose
1. The cleaned csv from data preparation stage is loaded into the pandas dataframe.
2. A new dataframe column is created to match the mp3 file names in the data, after a bit of cleaning.
3. The wavelet spectographs of one bird are observed across different qualities, to have a visual representation of the quality
It is evident that, the Qualities marked as A and B are useful in terms of model training.
4. The high quality sound files are selected, and split as train, test and validation datasets.
Librosa is used to extract the features of each file and stored in dataframe.
5. Deep learning model is trained. The hyper parameters are tuned until a satisfactory level of accuracy on train and test data are obtained.
Hyper parameters tuning:
Accuracy is observed for the model trained.
6. Once happy with the model, the birds in the given test data set are observed.
Conclusion
- It is observed that 4 out of the 15 test birds are Rose-Crested-Blue-Pipits. The rest are as follows:
