Detection of targets in remote-sensing data is a fundamental tool for many remote-sensing applications. False alarms generated by detectors are a common problem that reduces the value of remote sensing in general. A solution to the false alarm problem involves the implementation of Constant False Alarm Rate (CFAR) detection techniques that vary the detection threshold as a function of the sensed environment. The challenge of developing CFAR detectors is that if the alarm rate is set too high, there are a greater number of false hits, while if the alarm rate is set too low, there are a greater number of missed targets. AUG Signals has made significant contributions in the area of automatic CFAR detection, having developed a number of cutting-edge technologies in the following areas below.

One of AUG Signals’ most important contributions to CFAR detection is multi-CFAR detector. Since no single CFAR detector performs optimally in all clutter environments, overall detection performance increases when the detection decision from distinct CFAR detectors are fused together. Multi-CFAR detector developed by AUG Signals is able to fuse the results of multiple complimentary CFAR detectors to increase the probability of detection while still keeping the probability of false alarm constant.



  • Enhanced detection performance
  • Fully adaptive to local background
  • Guaranteed false alarm rates to be below a user defined level
  • Takes advantage of different polarimetric transformations and CFAR detection technologies
  • Fast and can be implemented for near real time applications

Fusion Based Polarimetric Detection optimally combines polarimetric transformations and decompositions, clutter analysis, modeling, fusion and multi-Constant False Alarm Rate (CFAR) detection to produce final detection results. The key technical attributes of this technique are the system’s ability to employ multiple CFAR detectors and use advanced clutter modeling technique. The clutter modeling technique is fully adaptable to background and can model more general clutter statistics. The multi-CFAR detector enables this algorithm to incorporate and best utilize the complementary capabilities of different single CFAR detectors.



  • Fully adaptive to background
  • Guarantees the false alarm rates to be below user defined levels
  • Takes advantage of different polarimetric transformations and CFAR detection technologies
  • Fast and can be implemented for near real time process
  • Enhanced detection performance

Figure 1 illustrates the difference between detection results (probability of detection versus Signal to Noise Ratio (SNR) for a fixed false alarm rate of 10-5) of Radarsat-2 polarimetric data between the novel polarimetric detection system employing fusion, the state-of-the-art technology available on the market today, and the most widely used commercial techniques for vessel detection.

Conventional CFAR detectors model the clutter and the target only with a probability distribution function (pdf); this includes the Fusion Based Polarimetric Detection System, mentioned above. With the availability of higher-resolution data, target objects described with multiple pixels will have the necessary information content for Markov chains modeling.

Markov Chain CFAR Detection models both background clutter and targets using Markov chains, providing information on the correlation between different sea states (in the case of ship detection) or other clutter environments. The algorithm is based on the likelihood ratio of the transitions to the observed value. Probability of detection and probability of false alarm will be derived using clutter and target transition matrices, taking into consideration the relative position of the matrices.

The computation complexity of the Markov chain CFAR detector depends on the number of neighbouring pixels used for Markov chain CFAR detection. The simplest case of Markov chain detector occurs when using only one test pixel, which equals the conventional PDF detector (the computational complexity (CC) is equal to the CC of the conventional PDF detector).

With more neighboring test pixels used for detection, the complexity of the algorithm is accordingly increased. The computational complexity of Markov chain detector is higher than the conventional PDF detector, and depends on the number of the test pixels used. In AUG Signals’ experiments, the Markov chain CFAR detector with 3, 5, and 9 test pixels runs about 2.66, 4.52, and 6.62 times respectively longer than the conventional PDF detector.


  • Significantly enhanced detection capability due to the new information employed – spatial correlation, which has never been used before, with slightly longer processing time as a trade-off
  • Fully adaptive to background
  • Guarantees the false alarm rates being below user defined levels
  • Very wide applicability, could be used for any data type, not limited to polarimetric SAR

AUG Signals’ Markov Chain CFAR detector has been successfully applied to Convair580 and Radarsat-2 polarimetric data (with spatial resolution of 3 meters) for vessel detection. Since the additional correlation information is employed, the Markov Chain CFAR detector results in advancing the performance of conventional CFAR detectors for targets with size equal or larger than 3 pixels. For point targets, the Markov Chain CFAR detector will be simplified to a conventional CFAR detector.

Figure 3 provides the detections derived by applying Markov Chain detector to the image presented in Figure 2. Figure 4 shows the performance improvements of using Markov Chain CFAR detector compared to the conventional PDF detector for vessel detection application using Radarsat-2 polarimetric data. Table 1 compares the probability of detection of using conventional PDF detector and Markov Chain detectors with 3, 5, and 9 pixels for all four polarization bands. The more pixels are used for Markov chain modeling, the better the performance is, but as a trade-off, longer processing time is required.

Figure 2. Radarsat-2 image for testing               Figure 3. Detections using Markov Chain Detector for the image in Figure 3

Figure 4. Vessel detection results comparison using Radarsat-2 polarimetric images

Table 1. Measured detection rate of Markov chain CFAR detector on 4 polarizations

AUG Signals’ fusion-based polarimetric detector can detect point targets. Its clutter modeling technique is fully adaptable to background (independent of sea condition) and can model more general clutter statistics. Therefore, whatever the sea condition is, optimal detection results can always be yielded. But as sea becomes rougher, the detectability will be reduced (which is similar to conventional CFAR detector). As mentioned before, the advantage of using Markov Chain CFAR detector is employing the correlation information. Therefore, various models could be generated to differentiate targets and different sea conditions, which enable the Markov Chain detector to better detect targets with 3 or more pixels in rough sea conditions. The more pixels are used in Markov Chain detector, the better performance is expected, especially in rough sea conditions.

The model based CFAR detector employs a precise statistical model to fit the background clutter. AUG Signals has unique techniques in modeling high resolution sea clutter using K and Generalized Compound (GC) distributions. Since the K-distribution is a mixture PDF, the parameters can be recursively estimated using expectation-maximization (EM) algorithm. Alternatively, a Generalized Compound (GC) model can be used to describe the deviations of the radar clutter statistics from the K-distribution. Our CFAR detection technology employing either K-PDF or GC-PDF outperforms the widely used distribution-free CFAR technology by at least 10%.
The Optimal Projection Adaptive Matched Subspace Detector detects the presence of signature pertaining to a particular material in hyperspectral signature of mixed samples. It performs optimal projection to emphasize the signature of the target material followed by matched subspace clutter suppression. The final detection result is calculated by transferring the generated abundance image into AUG Signals’ multi-CFAR detector, which includes several different single CFAR detectors and employs the Generalized Gamma model, whose parameters are calculated using a multidimensional minimization method (i.e., genetic algorithm).