Detect aircrafts in an AOI
Table of Contents
| Status | Operational |
| Launch year | Late 2021 |
| Latest release | Version 2, late 2022 |
| Data source | SAR |
| Delivery formats | JSON |
| Delivery methods | API, S3 |
What is the Product?
Ursa’s Aircraft Detection product is an operational, SAR-based solution for detecting and classifying aircrafts within user-specified areas of interest such as airfields and airports.
This product can be generated from spotlight imagery with accuracy of results varying based on the details of the SAR image acquisition mode (e.g., ground plane resolution).
Currently, imagery from the following SAR data providers are compatible with Ursa’s Aircraft Detection processor:
-
Capella
-
ICEYE
-
Airbus (coming soon)
-
Umbra (coming soon)
By default, aircraft are classified into the following 7 categories:
-
Fixed Wing, Fighter (Military)
-
Fixed Wing, Bomber (Military)
-
Fixed Wing, Heavy Lift (Military)
-
Fixed Wing, Medium Lift (Military)
-
Fixed Wing, Other ((Military)
-
Fixed Wing, Civilian
-
Rotary Wing
If other classification categories are required, please contact your Solutioner or Sales Engineer.
The Aircraft Detection product detects stationary aircraft only. Aircraft in flight, even those within the SAR image scene extent, will not be detected.
Performance Metrics
Accuracy Metrics
Accuracy performance is quantified using the following metrics.
-
Recall: True Positives / (True Positives + False Negatives)
-
Also referred to as Probability of Detection (Pd)
-
-
Precision: True Positives / (True Positives + False Positives)
-
This is complementary to the False Alarm Rate (FAR). (Precision = 1 - FAR)
-
-
F1: 2 * Precision * Recall / (Precision + Recall)
-
Common target recognition metric that combines precision and recall
-
-
Mean Average Precision (mAP): The mean (over all categories) of the Average Precision
Below, we provide accuracy metrics for our Aircraft Detection product when run on imagery containing targets, scenery, and vendor modes for which the detection model had previously been trained. These metrics represent expected performance for similar scenarios once adequate training data has been compiled. In the meantime, performance is expected to be worse as the model adapts to new imagery.
| Metric | Value |
| Recall | 78% |
| Precision | 86% |
| Ff1 | 82% |
| mAP | 81% |
Product Generation Latency Metrics
The following metrics represent approximate processing latency estimates.
-
Time from image capture to Ursa acquisition: 4-12 hours (varies)
-
Time from Ursa acquisition to product delivery: 10-15 minutes
-
Time from automated results to quality controlled results: 24-48 business hours (varies)
Standard Output File Formats
A standard Aircraft Detection product will include the following files:
|
File Name |
File Type |
Description |
|---|---|---|
|
aircraft_detections.json |
Aircraft Detections GeoJSON File |
GeoJSON file containing one feature for every detected aircraft, as well as supplemental information about the SAR collection and the algorithms used. Georeferencing is provided using a WGS 84 projection. |
|
chips/{chip_id}.tif |
TIFF/PNG |
One image subset (a.k.a. “chip”) is provided corresponding to each detection within the image. These files are available as Geotiff or PNG. These files are contained within a chips/ directory within the standard output directory structure. The sarChipIdentifier field within aircraft_detections.json provides a link to the correct image chip. |
Aircraft Detections GeoJSON File
The metadata and property fields provided within Aircraft Detections GeoJSON file are found in the following documents.
Ursa Analytic: GeoJSON Product Information
Object Detection: Product Format
The Aircraft Detection product uses the following feature properties to annotate aircraft detections. These properties are described in the Object Detection Product Format .
-
uuid
-
geometrySource
-
detectionNumber
-
sarLatitude
-
sarLongitude
-
sarDetectionTimeUTC
-
classificationCategory
-
classificationType
-
classificationSubtype
-
classificationClass
-
classificationModel
-
classificationProbability
If classification details beyond a description of the single most likely classification category are required, the classificationDetails field may also be provided.
Example Product Output
{
"type": "FeatureCollection",
"product": {
"productType": "Object Detection",
"documentVersion": "0.1"
},
"processors": [
{
"processorName": "YoloObjectDetector",
"version": "target_counting.4.2.1.best-20220206.pt",
"parameters": {
"aoi": {
"name": "Liangcheng",
"country_code": "CHN",
"polygons": [
[
[
25.686920658000076,
116.75070928900004,
375.684879765391
],
[
25.684789142000056,
116.74986980100005,
381.9599593085194
],
[
25.682277033000048,
116.74872678100007,
380.13125863742414
],
.......
.......
[
25.686920658000076,
116.75070928900004,
375.684879765391
]
]
]
},
"aoi_dilation_m": 0.0,
"distance_offset_m": 0,
"bearing_offset_deg": 0,
"object_detector_params": {
"model_name": "best-20220206.pt",
"num_input_rows": 640,
"num_input_columns": 640,
"tile_overlap": 0.3,
"nms_confidence_threshold": 0.4,
"nms_iou_threshold": 0.2
}
}
}
],
"sensorData": [
{
"sensorDataIndex": 0,
"sensorName": "ICEYE-X7",
"sensorType": "SAR",
"sensorMode": "SPOTLIGHT: SpotlightHigh",
"productType": "SLC",
"vendorId": "ICEYE_X7_SLC_SLH_511596_20220425T050606",
"startTimeUTC": "2022-04-25T05:06:06.549562Z",
"geometryReference": "POINT Z (116.74470520019531 25.675981521606445 390.6380310058594)",
"azimuthAngle": 76.69945997544089,
"incidenceAngle": 31.66148975914988,
"polarization": "VV",
"footprint": "POLYGON ((116.76674355866383 25.70576610206045, 116.77915977330761 25.658247786106458, 116.72287164431957 25.64623979206334, 116.7104356880881 25.69375367727714, 116.76674355866383 25.70576610206045))",
"ursaId": "be20551f-6aa7-4227-88be-d408bc042f52"
}
],
"features": [
{
"type": "Feature",
"geometry": {
"type": "Polygon",
"coordinates": [
[
[
116.74967263853739,
25.67891409566529
],
[
116.74955709263085,
25.67891409566529
],
[
116.74955709263085,
25.678711446860397
],
[
116.74967263853739,
25.678711446860397
],
[
116.74967263853739,
25.67891409566529
]
]
]
},
"properties": {
"uuid": "6d500eb2-4db6-4cac-a399-b5c13bdad839",
"geometrySource": "SAR",
"detectionNumber": 0,
"classificationClass": "Fighter",
"classificationProbability": 0.9001199007034302,
"classificationCategory": "Unknown",
"classificationModel": "Unknown",
"classificationSubtype": "Unknown",
"classificationType": "Unknown",
"sarDetectionTimeUTC": "2022-12-06T18:00:58.112455",
"sarLatitude": 25.678812771262844,
"sarLongitude": 116.74961486558412
}
},
{
"type": "Feature",
"geometry": {
"type": "Polygon",
"coordinates": [
[
[
116.74994876983546,
25.678936962711774
],
[
116.74981985031827,
25.678936962711774
],
[
116.74981985031827,
25.67872519359987
],
[
116.74994876983546,
25.67872519359987
],
[
116.74994876983546,
25.678936962711774
]
]
]
},
"properties": {
"uuid": "6d500eb2-4db6-4cac-a399-b5c13bdad839",
"geometrySource": "SAR",
"detectionNumber": 1,
"classificationClass": "Fighter",
"classificationProbability": 0.8988022804260254,
"classificationCategory": "Unknown",
"classificationModel": "Unknown",
"classificationSubtype": "Unknown",
"classificationType": "Unknown",
"sarDetectionTimeUTC": "2022-12-06T18:00:58.112455",
"sarLatitude": 25.678831078155824,
"sarLongitude": 116.74988431007687
}
},
{
"type": "Feature",
"geometry": {
"type": "Polygon",
"coordinates": [
[
[
116.75028176169148,
25.68017319821074
],
[
116.75015938311398,
25.68017319821074
],
[
116.75015938311398,
25.679959240660434
],
[
116.75028176169148,
25.679959240660434
],
[
116.75028176169148,
25.68017319821074
]
]
]
},
"properties": {
"uuid": "6d500eb2-4db6-4cac-a399-b5c13bdad839",
"geometrySource": "SAR",
"detectionNumber": 2,
"classificationClass": "Fighter",
"classificationProbability": 0.8961124420166016,
"classificationCategory": "Unknown",
"classificationModel": "Unknown",
"classificationSubtype": "Unknown",
"classificationType": "Unknown",
"sarDetectionTimeUTC": "2022-12-06T18:00:58.112455",
"sarLatitude": 25.680066219435588,
"sarLongitude": 116.75022057240272}
},
]
}