Stand Alone Multiple Anomaly Technique (SMART)

The Standalone Multiple Anomaly Recognition Technique (SMART) is a software tool that assists operators in the detection of contraband in cargo containers in the operating real world environment. For nearly two decades the federal government has vigorously supported improved technologies to counter use of the global supply chain for illegal or terrorist purposes without impeding the flow of commerce. The more than 50,000 cargo containers entering the land and seaports of the United States daily clearly provides terrorists and criminals with an attractive path for smuggling destructive weapons including weapons of mass destruction or illegal materials such as explosives, drugs, money or counterfeit items, into our nation. By developing and deploying non intrusive inspection (NII) cargo scanners (both x-ray and gamma ray) that enable operators to acquire and analyze up to 20 cargo images per hour a burden has been put on the operator to maintain a very high probability of contraband detection with a fractional level of false positives.

smart1
Figure 1 – SMART results on modeled cargo images, a) image A, b) image B, c) image C, d) image D. The anomalies detected in images A, B, C, and D are shown in figures e, f, g, and h, respectively

Creative Electron, Inc. (CEI) is addressing the need to provide operators with a capability to nominate potential contraband that are contained in the cargo conveyances. SMART is a stand-alone software package that can receive the x-ray data as files over standard computer interfaces and output the results as files over the same hardware interface. Threat materials can take many shapes and therefore an emphasis is placed on anomaly detection instead of specific shape detection. Current state of the art in detection of contraband relies on human operators that can result in a high false positive rate. The goal of this program is to facilitate improvements in human operator performance in the detection of contraband and to reduce the false positive rate of the overall process (human operator and signal processing combined). The other objective is for SMART to greatly reduce the time required to examine a truck or cargo conveyance image.
The anomaly recognition technique described in this paper relies on the use of sub-band decomposition of the NII image. In the case of a cargo container, regions of the space (or the entire container) are often filled with similar shaped objects. Thus, SMART takes advantage of the sub-band decomposition of the NII image to establish the parameters in the consistent content region(s) of the cargo that provide a baseline for anomaly detection. Based on this information, SMART scans the consistent image region (or the entire container) for anomalies. These anomalies are detected by SMART as deviations in the energy content in the sub-band decomposition of the image.
SMART measures deviations from the most common shape in the image using a frequency decomposition approach. Thus, SMART intrinsically filters the images in the frequency domain. As a consequence, SMART can handle relatively low quality (poor resolution or low contrast sensitivity) input images. This is a key advantage over standard techniques (e.g., automatic target recognition) that look for specific shapes in the NII image.
As an example of SMART’s performance, we ran the algorithm with images that represent a reasonably complex image from a simulated container. The technique used in these examples relies on statistical information from the image to determine the optimal threshold. Figure 1a to Figure 1d show preliminary modeled container images. In this example, the container should be filled with striped circles. Thus, everything that is not a striped circle must be identified as an anomaly. To demonstrate the potential of this concept, the anomalies in the figures vary in shape, color, and size. We also included one figure without any anomalies to assure us that the approach was not producing phantom targets.

The results in Figure 1e to Figure 1f correspond to the images in Figure1a to Figure 1d. The dark objects in Figure 1e, for example, correctly correspond to the location of the 3 anomalies present in Figure 1a. It is important to note that this determination was done automatically, without any human intervention. Similarly, the anomaly-free Figure 1c resulted in no targets for Figure 1g.
The anomalies in Figure 1 can easily be identified visually. Thus, in Figure 2 we modeled a similar container image with the picture of an AK47 as the anomaly to be detected (Figure 2a and Figure 2c). The transparency in Figure 2a is such that the position of the AK47 is easily identified visually. However, in Figure 2c the position of the AK47 is hardly perceived visually. Despite the changes in the transparency of the input images, SMART was able to correctly recognize the anomaly and position it in the image in Figure 2b (for the image in Figure 2a) and Figure 2d (for the image in Figure 2c).