Knowledge Discovery Lab

Craters

Theme:  

Crater Detection

Number of Instances:

Variable

Area:

Planetary, Object Detection

Attribute Characteristics:

Integer

Number of Attributes:

1090

Date Created

2011

Associated Tasks:

Classification

Missing Values?

No

 

Cropped craters from dataset source

 

 

1_24
Instances 6708
Attributes 1090

 

1_25
Instances 2935
Attributes 1090

 

2_24
Instances 578
Attributes 1090

 

2_25
Instances 603
Attributes 1090

 

3_24
Instances 496
Attributes 1090

 

3_25
Instances 727
Attributes 1090

Dataset Objective:
Determine if the instance is a crater or not a crater. 1=Crater, 0=Not Crater

Data Set Information:
This dataset was generated using HRSC nadir panchromatic image h0905_0000 taken by the Mars Express spacecraft. The images is located in the Xanthe Terra, centered on Nanedi Vallis and covers mostly Noachian terrain on Mars. The image had a resolution of 12.5 meters/pixel.

Data Set Generation:

Using the technique described by L. Bandeira (Bandeira, Ding, Stepinski. 2010.Automatic Detection of Sub-km Craters Using Shape and Texture Information) we identify crater candidates in the image using the pipeline depicted in the figure below. Each crater candidate image block is normalized to a standard scale of 48 pixels. Each of the nine kinds of image masks probes the normalized image block in four different scales of 12 pixels, 24 pixels, 36 pixels, and 48 pixels, with a step of a third of the mask size (meaning 2/3 overlap). We totally extract 1,090 Haar-like attributes using nine types of masks as the attribute vectors to represent each crater candidate.
The dataset was converted to the Weka ARFF format by Joseph Paul Cohen in 2012.

Attribute Information:
We construct a attribute vector for each crater candidate using Haar-like attributes described by Papageorgiou 1998. These attributes are simple texture attributes which are calculated using Haar-like image masks that were used by Viola in 2004 for face detection consisting only black and white sectors. The value of a attribute is the
difference between the sum of gray pixel values located within the black sector and the white sector of an image mask. The figure below shows nine image masks used in our case study. The first five masks focus on capturing
diagonal texture gradient changes while the remaining four masks on horizontal or vertical textures.

Download: https://github.com/ieee8023/CraterDataset

This crater dataset is used in the following papers:

W. Ding, T. Stepinski, Y. Mu*, L. Bandeira*, R. Vilalta, Y. Wu, Z. Lu*, T. Cao*, X. Wu, “Sub-Kilometer
Crater Discovery with Boosting and Transfer Learning”, ACM Transactions on Intelligent Systems and
Technology, Vol. 2, Issue 4, July, 2011. (PDF URL: http://www.cs.umb.edu/~ding/papers/tist11.pdf)

Liu, Siyi, Wei Ding, Feng Gao, and Tomasz F. Stepinski. “Adaptive Selective Learning for Automatic Identification of Sub-kilometer Craters.” Neurocomputing 92, no. 0 (September 1, 2012): 78–87.
PDF

Liu, Siyi, Wei Ding, Joseph Paul Cohen, Dan Simovici, and Tomasz F. Stepinski. “Bernoulli Trials Based Feature Selection for Crater Detection.” In Proceedings of the 19th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, pp. Chicago, Illinois: ACM, 2011.
PDF

 

Citations:

W. Ding, T. Stepinski, Y. Mu*, L. Bandeira*, R. Vilalta, Y. Wu, Z. Lu*, T. Cao*, X. Wu, “Sub-Kilometer
Crater Discovery with Boosting and Transfer Learning”, ACM Transactions on Intelligent Systems and
Technology, Vol. 2, Issue 4, July, 2011. (PDF URL: http://www.cs.umb.edu/~ding/papers/tist11.pdf)

Bandeira, L., W. Ding, and T. F Stepinski. “Automatic Detection of Sub-km Craters Using Shape and Texture Information.” In Proceedings of the 41st Lunar and Planetary Science Conference. The Woodlands, Texas, 2010. http://adsabs.harvard.edu/abs/2010LPI....41.1144B.
PDF

Comments (0) Trackbacks (0)

No comments yet.


Leave a comment

You must be logged in to post a comment.

Trackbacks are disabled.