[cs461][COAL] Feature extraction

[Clayton, Heidi Ann]

I'm writing about what method we will probably use for feature extraction of the AVIRIS data. Neural networks seem like the best choice since, to my limited knowledge, they are good for data that has many dimensions. I need to produce two alternatives that wouldn't be ideal for this, and I'm coming up blank. Does anyone know any feature extraction methods that they don't like for this?

Thanks,

Heidi Clayton

[Clayton, Heidi Ann]

Also, will we be preprocessing the AVIRIS data?

[Lewis John Mcgibbney]

Hi Heidi,

Thanks for approaching this rather tricky and difficult topic. It is my no means a walk in the park so thank you. Replies inline.

On Sun, Nov 13, 2016 at 11:07 PM, Clayton, Heidi Ann <clay...@oregonstate.edu> wrote:

I'm writing about what method we will probably use for feature extraction of the AVIRIS data.

Great. I have also been thinking about this a lot recently and as a matter of fact I was at a guest lecture here @JPL on extactly this topic. The lecture basically stated that 'state-of-the-art' is as follows

Active Learning for Supervised Classification of Hyperspectral Remote Sensing Data

 

Presented by Melba Crawford

Purdue University

 

Abstract

Accurate land cover classification that ensures robust mapping under diverse conditions is important in environmental studies where the identification of the land cover changes and its quantification have critical implications for management practices, ecosystem health, and the impact of climate change.  Hyperspectral data provide enhanced capability for more accurate discrimination of land cover, but significant challenges remain for supervised classification, including highly correlated spectral bands, high dimensionality, and nonlinear spectral response.  Advanced methods in machine learning, including active learning that focuses on developing informative training sets with minimal redundancy, have been demonstrated to promote greater exploitation of the information in both labeled and unlabeled data, while significantly reducing the cost of data collection.  Recent developments in active learning for classification of hyperspectral remotely sensed data, in combination with feature extraction via metric learning, are outlined and demonstrated using airborne and space-based hyperspectral data.

 

Neural networks seem like the best choice since, to my limited knowledge, they are good for data that has many dimensions.

Well, lets keep in mind that the paper I provided was from 1995 and some things have moved on. I think that neural network-based model approached are going to be valuable. There is no doubt about that. I would like to get my hands on the slides provided by Prof. Melba as above and we can discuss as a group.

 

I need to produce two alternatives that wouldn't be ideal for this, and I'm coming up blank. Does anyone know any feature extraction methods that they don't like for this?

It looks like you came up with a few and they have been included with your current preference being NN. That is fine and was can refine over time. Let's discuss on Thursday.

Thanks

[EvandroCT]

Hi folks,

I'm the student assigned to this GSoC/ESIP project mentored by Lewis and related to COAL project. I've seen this topic while walking around the mailing list as a GSoC's Community Bounding activity and though it would be a good idea to suggest an algorithm that has been proven to efficiently reduce dimensionality of hyperspectral data. The algorithm is called SPA (Successive Projections Algorithm) and I've tested it by myself over spectrophotograms of petrochemical samples in order to calibrate a quality model. The work was done as the final term project of the Patter Recognition course taken at my masters degree program. The project's paper is attached. Below, the translated abstract:

This paper describes a statistical technique known as 'Multivariate Calibration', whose objective is the construction of a predictive model capable of estimating one or more quantities based on measured values of a set of explanatory variables. For this, a variable selection algorithm known as 'Successive Projection Algorithm' (SPA) have been used over a high dimensional petrochemical database. The application of the algorithm allowed a reduction of 99.3% of the total of explanatory variables besides minimizing the error rate of prediction of the model.

Following there's some papers related to SPA:

• https://www.sciencedirect.com/science/article/pii/S0169743901001198
  
• https://www.researchgate.net/publication/229349059_A_Variable_Elimination_Method_to_Improve_the_Parsimony_of_MLR_Models_Using_the_Successive_Projections_Algorithm
  

Please, feel free to request more information.

Cheers,

Evandro C Taquary

[Lewis John Mcgibbney]

Hi Evandro,

Thanks for posting this. Do you happen to have a translation of the attachment?

Thanks

Lewis

--
You received this message because you are subscribed to the Google Groups "COAL - Coal and Open-pit surface mining impacts on American Lands" group.
To unsubscribe from this group and stop receiving emails from it, send an email to coal-capstone+unsubscribe@googlegroups.com.
To post to this group, send email to coal-c...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/coal-capstone/501bfa98-155c-4725-973d-3839cc52ba74%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Lewis

Dr. Lewis J. McGibbney Ph.D, B.Sc

Skype: lewis.john.mcgibbney

[EvandroCT]

Lewis, I haven't a translation, unfortunately. But if you need any details or even the dataset and scripts used, I can provide them if you wish. Although this isn't a exactly a feature extraction approach, I think that, by dealing the high dimensionality, it fits some of the project's needs.

Em terça-feira, 8 de maio de 2018 16:27:28 UTC-3, Lewis John Mcgibbney escreveu:

Hi Evandro,

Thanks for posting this. Do you happen to have a translation of the attachment?

Thanks

Lewis

On Sat, May 5, 2018 at 11:56 AM, EvandroCT <evandro...@gmail.com> wrote:

Hi folks,

I'm the student assigned to this GSoC/ESIP project mentored by Lewis and related to COAL project. I've seen this topic while walking around the mailing list as a GSoC's Community Bounding activity and though it would be a good idea to suggest an algorithm that has been proven to efficiently reduce dimensionality of hyperspectral data. The algorithm is called SPA (Successive Projections Algorithm) and I've tested it by myself over spectrophotograms of petrochemical samples in order to calibrate a quality model. The work was done as the final term project of the Patter Recognition course taken at my masters degree program. The project's paper is attached. Below, the translated abstract:

This paper describes a statistical technique known as 'Multivariate Calibration', whose objective is the construction of a predictive model capable of estimating one or more quantities based on measured values of a set of explanatory variables. For this, a variable selection algorithm known as 'Successive Projection Algorithm' (SPA) have been used over a high dimensional petrochemical database. The application of the algorithm allowed a reduction of 99.3% of the total of explanatory variables besides minimizing the error rate of prediction of the model.

Following there's some papers related to SPA:

• https://www.sciencedirect.com/science/article/pii/S0169743901001198
  
• https://www.researchgate.net/publication/229349059_A_Variable_Elimination_Method_to_Improve_the_Parsimony_of_MLR_Models_Using_the_Successive_Projections_Algorithm
  

Please, feel free to request more information.

Cheers,

Evandro C Taquary

Em segunda-feira, 14 de novembro de 2016 05:07:21 UTC-2, claytonh escreveu:

I'm writing about what method we will probably use for feature extraction of the AVIRIS data. Neural networks seem like the best choice since, to my limited knowledge, they are good for data that has many dimensions. I need to produce two alternatives that wouldn't be ideal for this, and I'm coming up blank. Does anyone know any feature extraction methods that they don't like for this?

Thanks,

Heidi Clayton

--
You received this message because you are subscribed to the Google Groups "COAL - Coal and Open-pit surface mining impacts on American Lands" group.

To unsubscribe from this group and stop receiving emails from it, send an email to coal-capston...@googlegroups.com.

To post to this group, send email to coal-c...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/coal-capstone/501bfa98-155c-4725-973d-3839cc52ba74%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Lewis John Mcgibbney]

Replies inline

On Wed, May 9, 2018 at 4:50 PM, EvandroCT <evandro...@gmail.com> wrote:

Lewis, I haven't a translation, unfortunately. But if you need any details or even the dataset and scripts used, I can provide them if you wish.

Thanks

 

Although this isn't a exactly a feature extraction approach, I think that, by dealing the high dimensionality, it fits some of the project's needs.

I agree. I'm working my way through the other resources in the meantime.

Thanks

 

To unsubscribe from this group and stop receiving emails from it, send an email to coal-capstone+unsubscribe@googlegroups.com.

To post to this group, send email to coal-c...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/coal-capstone/4eb22d32-bd43-4627-8f4f-660ab21b541c%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.