Contest No. 4

Damaged Building Detection Using Multi-Temporal High Resolution Google Earth Images

Description:

After an earthquake, the localization and detailed assessment of damaged buildings is very important in order to guide response operations, recovery tasks, or for insurance purposes. Several very high spatial resolution (VHR) remote sensing platforms and sensors are used for these purposes. However, these VHR images was commercial and purchasing the time series of these images is costly. Also, there is an overall interest in the use of automated methods to perform these tasks, using free available multi-temporal Google Earth (GE) images. So the main goal of this contest is to detect the damaged building after earthquake using multi-temporal VHR Google Earth images. Two VHR Google Earth images were used in this contest including acquired images before and after the earthquake (for detecting the damaged building). Accordingly, the individual steps of the data processing could be as follows:

• Preprocessing of two Google Earth (GE) images such as fine co-registration, multi-temporal radiometric normalization, building shadow removal.
• Masking the no-building area.
• Optimum feature extraction/selection for building detection.
• Segmentation and extraction of buildings for two images and provide two vector maps.
• Comparing the two segmented images and detecting the damaged building, and finally presenting the results as shape file format
• Writing a technical report by describing all operational individual steps

Dataset:

On 12 November 2017 at 18:18 UTC, an earthquake with a moment magnitude of 7.3 occurred on the Iran–Iraq border with the dominated places of Kermanshah Province in Iran. The study area of this contest is Sarpol-e Zahab city in Kermanshah Province (figure 1).

Figure 1. The study area of contest (Sarpol-e Zahab city in Kermanshah Province) (©Google Maps)

The VHR Google Earth images were used in this contest are available at this website. The preview of these multi-temporal image are shown in figure 2 and figure 3.

 Figure 2. Google earth image acquired from Sarpol-e Zahab city before the earthquake

 Figure 3. Google earth image acquired from Sarpol-e Zahab city after the earthquake

Deliverables:

Outputs included in this contest will be as follows:

• Technical report in docx or pdf describing all detail analysis (NOT more than 5000 words)
• Damaged building map in shape format (.shp)
• ^* Programming Codes or application software (if employed)

*Submission the applicable programming codes will have additional score for participants.

Evaluation Criteria:

The following criteria are considered for the evaluation of the accomplished tasks:

-          Precision of detected damaged buildings

-          Recall of detected damaged buildings

-          F1-score of detected damaged buildings

-          Area under the receiver operating characteristic Curve (ROC AUC) of detected damaged buildings

-          Geometric accuracy of the detected damaged buildings.

-          To develop and use your own program code (preferably open source) instead of using existing commercial tools.

-          A comprehensive technical report (in Persian or English).

-          Level of scientific innovation.

Data availability:

The web link for downloading the multi-temporal GE images will be available upon the completion of the registration form.