Detect Image Anomalies (Image Analyst Tools)

Summary

Processes a multiband or hyperspectral image and creates an anomaly score raster. An anomaly score raster is a single band raster, with values between 0 and 1.

Usage

An anomaly in an image refers to pixels that are significantly different from the background values, such as ships in the ocean, vehicles on a road, or human development in natural areas.  The tool supports the Reed-Xiaoli Detector (RXD), Uniform Target Detector (UTD), and KMEANS image anomaly detection methods.
The RXD option of the Anomaly Calculation Method parameter calculates the pixels' Mahalanobis distance to the background that is defined by the mean. It has the following formula:

\(δ _{RXD} (r)=(r-µ) ^{T} K ^{-1} _{LxL} (r-µ)\)

Where r is the sample pixel spectra, µ is the mean spectra, K is covariance, and L is the number of bands.
The UTD option of the Anomaly Calculation Method parameter is similar to the RXD option but extracts a background using a unit vector. It has the following formula:

\(δ _{UTD} (r)=(1-µ) ^{T} K ^{-1} _{LxL} (1-µ)\)

Which defines the anomaly by replacing (r-u) in the RXD method with (1-u).
The KMEANS option of the Anomaly Calculation Method parameter identifies pixels that deviate significantly from the established clusters in the data using K-means clustering algorithms.
The output is an anomaly score raster with decimal values between 0 and 1, where 0 is the background and large values approaching 1 are potential anomalies. The anomaly pixels can be filtered using raster functions, such as the Remap or Mask function. The entire anomaly detection workflow can be performed using the Anomaly Detection Wizard.
Use Dynamic Range Adjustment (DRA) in the output layer symbology to better visualize anomaly pixels.

Label	Explanation	Data type
Input Raster	A multiband or hyperspectral image.	Raster Dataset; Raster Layer; Mosaic Dataset; Mosaic Layer; Image Service
Output Raster	A single band raster that stores the anomaly scores between 0-1 as a floating point number. Zero (0) is the background value, and large values approaching 1 are potential anomaly pixels. Use the file extension to specify the output format, including `.tif` (TIFF), `.crf` (CRF), `.mrf` (MRF), and `.dat` (ENVI DAT).	Raster Dataset
Anomaly Calculation Method (Optional)	Specifies the anomaly calculation method that will be used. RXD—The RXD method will be used to extract pixels that are significantly different from the background pixel values. This is the default. UTD—The UTD method will be used to extract background pixels from the input image. KMEANS—The KMEANS method will be used to extract pixels that significantly deviate from the established clusters in the data using K-means clustering algorithms.	String
Number of Clusters (Optional)	The number of clusters that will be used when the Anomaly Calculation Method parameter is set to KMEANS.	Long
Background Region (Optional)	A polygon feature class that will define the region to be used to calculate background statistics when the Anomaly Calculation Method parameter is set to RXD or UTD.	Feature Set
Recompute Statistics (Optional)	Specifies whether statistics will be recomputed for the input raster when the Anomaly Calculation Method parameter is set to RXD or UTD. The RXD and UTD options require accurate statistics in which the skip factor must be 1 when statistics are computed. Checked—Statistics will be recomputed for the output score raster. This is the default. Unchecked—Statistics will not be recomputed for the output score raster.	Boolean

DetectImageAnomalies(in_raster, out_raster, {method}, {num_cluster}, {background_region}, {recompute_stats})

Name	Explanation	Data type
in_raster	A multiband or hyperspectral image.	Raster Dataset; Raster Layer; Mosaic Dataset; Mosaic Layer; Image Service
out_raster	A single band raster that stores the anomaly scores between 0-1 as a floating point number. Zero (0) is the background value, and large values approaching 1 are potential anomaly pixels. Use the file extension to specify the output format, including `.tif` (TIFF), `.crf` (CRF), `.mrf` (MRF), and `.dat` (ENVI DAT).	Raster Dataset
method (Optional)	Specifies the anomaly calculation method that will be used. `RXD`—The RXD method will be used to extract pixels that are significantly different from the background pixel values. This is the default. `UTD`—The UTD method will be used to extract background pixels from the input image. `KMEANS`—The KMEANS method will be used to extract pixels that significantly deviate from the established clusters in the data using K-means clustering algorithms.	String
num_cluster (Optional)	The number of clusters that will be used when the `method` parameter is set to `KMEANS`.	Long
background_region (Optional)	A polygon feature class that will define the region to be used to calculate background statistics when the `method` parameter is set to `RXD` or `UTD`.	Feature Set
recompute_stats (Optional)	Specifies whether statistics will be recomputed for the output score raster when the `method` parameter is set to `RXD` or `UTD`. The `RXD` and `UTD` options require accurate statistics in which the skip factor must be 1 when statistics are computed. `RECOMPUTE_STATS`—Statistics will be recomputed for the output score raster. This is the default. `NOT_RECOMPUTE_STATS`—Statistics will not be recomputed for the output score raster.	Boolean

Code sample

DetectImageAnomalies example 1 (Python window)

This example processes a hyperspectral image using the RXD option and creates an anomaly score raster.

# Import system modules
import arcpy
from arcpy.ia import *

tempanomaly = arcpy.ia.GenerateMultidimensionalAnomaly(
    "c:/data/climateData.nc", "temperature", "DIFFERENCE_FROM_MEAN",
    "ALL", "DATA", None)

tempanomaly.save("c:/data/TempAnomaly.crf")

DetectImageAnomalies example 2 (stand-alone script)

This example processes a hyperspectral image using the KMEANS option and creates an anomaly score raster.

# Import system modules
from arcpy.ia import *

#Define variables
input_image = r"c:\data\hsi_image.tif"
num_of_cluster = 1

anomaly_raster = arcpy.ia.DetectImageAnomalies(
    in_raster=input_image,
    method="KMEANS",
    num_cluster=num_of_cluster,
    recompute_stats="RECOMPUTE_STATS"
)
anomaly_raster.save(r"c:/test/anomaly_score_raster.tif")

Environments

Cell Alignment, Cell Size, Compression, Current Workspace, Geographic Transformations, NoData, Output CONFIG Keyword, Output Coordinate System, Extent, Pyramid, Raster Statistics, Resampling Method, Scratch Workspace, Snap Raster, Tile Size

Licensing information