# Kraken Output Formats¶

## Output GeoJSON of Detection Algorithms¶

GeoJSON with a FeatureCollection where each Feature represents an individual detection. All features have geometry of type Polygon and properties which are specific to algorithm which this is result of.

geometry of each region is its minimum-area bounding rectangle.

Note

Minimum-area bounding rectangles can have any orientation, i.e. they are not necessarily aligned with parallels and meridians or the feature’s orientation property.

Note

Detections are included if their centroid lies in requested area. This removes the need for detection de-duplication when the tiles are aggregated on client side.

{
"type": "FeatureCollection",
"features": [
{
"type": "Feature",
"properties": {
"class": "cars",
"area": 7.6,
"count": 1,
"orientation": -28.5,
"aspectRatio": 0.22,
"confidence": 0.95
},
"geometry": {
"type": "Polygon",
"coordinates": [
[
[153.12104687732085, -27.388643101636596],
[153.12137530696452, -27.389043532193686],
[153.12076821186633, -27.389436088316238],
[153.12043978372787, -27.389035658966332],
[153.12104687732085, -27.388643101636596]
]
]
}
}
]
}


### Object Detection Feature Properties¶

• area – area of detected object in $$m^2$$; always present.
• areaPercentage – ratio of the object area and the area of its bounding box; may be unavailable.
• aspectRatio – ratio between longer side and shorter side of the minimum-area enclosing oriented rectangle; may be unavailable.
• bboxArea – area of the object’s bounding box in $$m^2$$; may be unavailable.
• class – determines object classification; always present.
• confidence – a number between 0 and 1; may be unavailable.
• count – number of objects represented by the Feature; always present.
• eccentricity - a float number between 0 and 1 (inclusive). It measures region linearity, where 1 means the region has very linear (eccentric) shape (e.g. narrow straight road). may be unavailable.
• latLonCenter – latitude and longitude of the region’s centroid; may be unavailable.
• max – maximum over pixel probabilities of the region; may be unavailable.
• mean – mean of region’s pixel probabilities; may be unavailable.
• median – median of region’s pixel probabilities; may be unavailable.
• min – minimum over pixel probabilities of the region; may be unavailable.
• orientation – angle between y-axis of an image and axis of a detected object given in degrees (counter-clockwise); present only if count is 1; may be unavailable.
• solidity – a float number between 0 and 1 (inclusive). It measures compactness, where 1 means the attention region has fully convex shape (e.g. circle); may be unavailable.
• std – standard deviation of region’s pixel probabilities; may be unavailable.
• shorterSide length of shorter side of the minimum-area enclosing oriented rectangle in meters may be unavailable.
• longerSide – length of longer side of the minimum-area enclosing oriented rectangle in meters may be unavailable.

## Attention Regions¶

Attention regions are areas within the analyzed extent where a potentially interesting change has happened. Attention regions GeoJSON has following properties:

• mean - Average value of the change heatmap inside the attention region; may be unavailable..
• solidity - a float number between 0 and 1 (inclusive). It measures compactness, where 1 means the attention region has fully convex shape (e.g. circle); may be unavailable..
• eccentricity - a float number between 0 and 1 (inclusive). It measures region linearity, where 1 means the attention region has very linear (eccentric) shape (e.g. narrow straight road); may be unavailable..

Metadata file contains one or more of following parts: areasM2, bandStatistics, analysisMetadata and normalizationStd depending on Map Type. For imagery and heat maps only bandStatistics are provided. For segmentation Map Types analysisMetadata and areasM2 are provided. For change heat maps bandStatistics and areasM2 are provided. normalizationStd is only provided in SAR change algorithms that have adaptiveThreshold.

Example of WRUNC Map Type:

{
"areasM2":
{
"water": 1161.74,
"urban": 0.0,
"nonurban": 2857.35,
"clouds": 0.0
},
{
"sScore": 1.22,
"algoVersion": "148"
}
}


### Area of Segmentation and Change Algorithms¶

Area calculated by segmentation algorithms contains square meters of area covered by each segmentation class.

Area calculated by change algorithms contains the area in square meters of significant change between the subsequent image acquisitions.

Example:

{
"areasM2":
{
"water": 1161.74,
"urban": 0.0,
"nonurban": 2857.35,
"clouds": 0.0
}
}


### Statistics for Band Pixel Intensities¶

Per band statistics computed from band pixel intensities (currently, only arithmetic mean). Note that mean might not be available if the analyzed area does not contain any valid data (i.e. analyzedAreaM2 is equal to zero).

Example:

{
"bandStatistics":
{
"heatmap":
{
"analyzedAreaM2": 932515.2,
"meanHeat": 0.013828165152889
}
}
}

• analyzedAreaM2 - Area of analyzed region in $$m^2$$.
• meanRadiance - Mean flux power per unit of solid angle (see Radiance and Reflectance) computed inside the analyzed area. Only available for datasets with radiance coefficients.
• meanHeat - Arithmetic mean of structural dissimilarity (in range $$[0, 1]$$) for change detection or normalized difference index, e.g. NDVI, (in range $$[-1,1]$$) computed inside the analyzed area. Available for all heatmap map types except for nl-change.
• meanAbsChange - Arithmetic mean of absolute values of relative change (in range $$[0, 1]$$). Available for nl-change.

Analysis Metadata has the following properties:

• sScore – s-score is a value between 0 and infinity that indicates suitability of imagery for analysis by the particular algorithm; s-score lower than 1 signifies imagery of superior quality. Values higher than 3.5 indicate imagery of inferior quality and less representative results.
• algoVersion – version of the algorithm that produced this analysis. Note that each algorithm (potentially on each dataset) is versioned independently.
• binarizationMethod – method determines process of converting probabilities into predictions. Possible choices are [“bepThreshold”, “argmax”, “argmaxBackground”, “adaptiveThreshold”, “fixedThreshold”].
• bepThreshold – break even point threshold. Must be in <0, 1> when binarizationMethod is set to “bepThreshold”, not specified otherwise.
• adaptiveThreshold – thresholding with respect to the scene statistics. Is a multiple of standard deviation. Must be of type float when binarizationMethod is set to “adaptiveThreshold”, not specified otherwise.
• fixedThreshold – threshold directly specified by value. Used when binarizationMethod is set to “fixedThreshold”, not specified otherwise.
• binarized – boolean flag indicating if the SKI went through binarization process.

Example:

{
{
"sScore": 1.22,
"algoVersion": "148",
"bepThreshold": 0.55,
"binarizationMethod": "bepThreshold"
}
}


### Normalization standard deviation¶

This metadatum is found only in metadata of SAR change algorithms that use adaptive threshold so that we are able to actually find out the threshold that was used for given analysis. It is mainly useful for tuning of normalization areas for analyses.

Example:

{
"normalizationStd": 0.123
}


## Segmentation SKI¶

Segmentation SKI files have a separate band for each detection class. The bands are named after the classes, for example band with cloud segmentation is called clouds. Bands are uint8 where “1” means that the particular pixel belongs to the class and “0” means the opposite.

Apart from geo-referencing metadata the SKI may also contain analysis metedata.

## Analysis GeoTIFF¶

Analysis GeoTIFFs can be used as a standard output of our algorithms. GeoTIFFs are created from our proprietary SKI file format.

## Visualization GeoTIFF¶

Visualizations are created as colored version of results of our algorithms, GeoTIFFs have 4 (RGBA) or 2 bands (grayscale, alpha) based on input imagery. JPEG lossy compression is used, thus the visualization GeoTIFFS should be used only for viewing purposes, not analytical ones.

## Polygon Statistics¶

Polygon Statistics file is a JSON with basic statistics computed on geometry per band. All statistics are computed only on valid pixels which are inside the given geometry. Basic statistics comprise of:

• pixelCount $$\sum_{t=1}^{N}1$$
• valueSum $$\sum_{t=1}^{N}x_{t}$$
• errorSquareSum $$\sum_{t=1}^{N} \|x_{t} - x_{mean}\|^2$$