www.radartutorial.eu www.radartutorial.eu Radar Basics

Coherent Change Detection

Figure 1: SAR image of an earlier time as a reference

Figure 1: SAR image of an earlier time as a reference

Figure 2: CCD image with the detected differences

Figure 2: CCD image with the detected differences

Coherent Change Detection

Coherent Change Detection (CCD) can be carried out in a synthetic aperture radar. This requires precise storage of raw data, exact image processing, and calculation. The method is used in civil and military applications and is a powerful technique for detecting changes on the earth's surface. Even the slightest changes on the earth's surface can be detected. A civil application is, for example, the evaluation of storm damage in forests.

Functional Principle

The procedure is similar to the well-known game of puzzle picture “Spot the difference”: an original image and a falsified copy are examined and all changes must be detected. Often the last change cannot be found and then only a technical procedure helps: both images are superimposed on each other and the differences become more apparent. But this is then actually only the Magnitude Change Detection procedure (or Non-Coherent Change Detection, NCCD), which only works for an optically captured image if the illumination situation in both images is comparable.

With radar the illumination situation is always comparable. In Coherent Change Detection, the two radar images are compared in amplitude and phase. (With radar, coherence always means a related phase shift to a reference oscillation.) The data source is thus a point in the chain of radar signal processing where the phase information is not yet lost: i.e. before demodulation! This means that much more accurate changes can be detected. Magnitude change detection is also possible with a SAR, but then after demodulation (see Pulse Integration)

In the coherent change detection image calculated by the SAR, even the car traces in the landscape can then be seen, simply because the blades of grass on the meadow were laid in a certain direction (direction of travel) as a result. The method is militarily usable because it recognizes primitive camouflage. If suddenly a tree is standing there, which was not there yesterday, then someone wants to hide something there! Such changes are highlighted in the image in a signal color (red or blue). Red means that an object has disappeared from the first image, blue means that an object has been added. A civil application is for example the evaluation of storm damage in forests.

Calculation of the image

Not only is the image compared pixel by pixel using cross-correlation, but its size is compared with the surroundings of this pixel, i.e. the average of a small area (for example 3×3 pixels). For each of these areas a coherence is calculated γ, which says how exactly this area of one image matches the area of the following image. If γ = 1, there is an exact match; if γ = 0, not even by pure randomness would something match. From a size of the deviation, for example γ < 0.9, a color change will occur.

For example, the lane on a lawn would still be visible after days and would cause a change to about γ = 0.96 … 0.98. In the case of lawn, a slight change in distance in the range of a few centimetres can be measured by pressing down the blades of grass. So the size of the echo signal hardly changes, but the phase position is completely different! Thus, an image which only compares the amplitudes (i.e. a non-coherent method) would not detect anything at all. A picture which normalizes the magnitude of these amplitudes and shows only this phase difference would mark such lanes particularly strongly. However, such a picture would look rather noisy and would also produce many false alarms, because for example the wind also influences the position of the leaves or blades of grass. For this reason, amplitude and phase must be compared at the same time, so that the image of the change is as contrasty as possible. Thresholds can also be defined for such an image and radar-typical parameters such as the probability of detection and false alarm rate (or probability of false alarms) can be calculated.