What is Passive Millimiter-Wave Imaging?

我们将毫米波（毫米波-MMW）称为波长在[10~1]毫米范围内，频率在[30~300]GHz范围内的电磁波。毫米波与雷达系统发射的波是同一种波，而且它们离微波区很近。毫米波技术因其穿透云、雾、雨或沙尘暴的能力而引起人们的兴趣[Yuj03]。事实上，衰减比可见光小大约一百万倍。无源毫米波（PMMW）摄像机不发射任何辐射。它们收集物体自然发射的辐射，或来自物体反射的其他来源的辐射。PMMW相机从场景的辐射中获得由其中的物体反射或发射的二维图。

We call millimeter waves (Millimeter Wave - MMW) to electromagnetic waves with wavelengths in [10, 1] mm range, and frequencies in [30,300] GHz range. MMW waves are the same kind of waves as the emitted, for example, by radar systems and they are near from microwave region. MMW technology generates interest due to its ability to penetrate clouds, fog, rain or sand storms [Yuj03]. In fact attenuation is about a million times smaller than visible light. Passive Millimeter Wave (PMMW) cameras do not emit any radiation. They collect radiation naturally emitted by objects, or coming from other sources that reflects in objects. A PMMW camera obtains a bidemensional map from radiation of the scene, reflected or emitted by objects therein.

Fig. 1 allows to compare a visible image (Fig. 1a) and a PMMW image (PMMWI) from the same scene (Fig. 1b). The car has a high reflectivity, its upper part reflects sky radiation (60 °K a 94 GHz [Yuj03]), while its lower part reflects radiation emitted by the ground (~300 °K). PMMWI from Fig. 1b has the typical characteristics from PMMWI, in terms of low resolution and noise. PMMW camera structure is similar to a photo camera. The main elements are the focusing system and the array of receivers, on which the image of the scene is formed (see Fig. 1c).

PMMW cameras used nowadays in surveillance and security have a single receiver [Gop10] (see Fig. 2a) or an array of receivers [Cal10] (see Fig. 2c) that, using a scanning system, allow to acquire PMMWI from the scene. Scene scan use to be slow, because of that Compressive Sensing (CS) techniques are being developing [Can08], to obtain images from the scene by using a single receiver and a coded aperture mask [Gop11,Bab11,Gop12] (see Fig. 2b). Fig. 2c shows a photo of a PMMW camera used in security purposes and developed by Alfa Imaging SA. This photo shows the mirror based scanning system, see scheme in Fig. 2d, that allows image acquisition by a 32 MMIC receivers array. On the right hand side of this paragraph, the camera developed by the company Wawecamm is shown.

MMW technology has several applications, however security is its main application. MMW technology allows to obtain information about hidden objects carried by persons that access to a public transport station or critical environments under terrorist threat. Any object, that prevents or hinders the MMW radiation coming from human body due to its temperature, could be detected in a PMMWI. Both methalic and non-methalic objects such as ceramic knives, ceramic weapons, liquids, dielectric explosive materials could be detected [Alx09,Alx10]. Non-methalic threats can not be detected by current security sytems. On the other hand, security control based on PMMW systems, does not emit any radiation so it can be used even with pregnant women.

Fig. 3a shows a PMMW camera developed by Alfa Imaging that integrates a camera like it is shown in Fig. 2c, with software and hardware elements developed by them to process PMMWI. Person from Fig. 3b hides an object under his vest. Fig. 3c shows the PMMWI and the output of the image processing algorithm that distinguish human silhouette and detects the hidden threat [Alx10]. Fig. 3d shows the image that appears in the operator screen, a visible image with the silhouette and the threat overlapping. Threats are marked over visible image to preserve privacy of individuals. Fusion techniques to overlap visible information and PMMWI is also necessary to facilitate operation tasks.

Research and development areas in this project

The research to be carried out in this project is considered relevant because the use of passive millimeter wave systems in security applications has a great potential. Despite of its potential, as noted above, the processing and extraction of semantic information from these images are research and development areas almost unexplored. This is mainly due to the proprietary nature of such images, obtained mainly in the corporate sector. In the project two members of Spanish companies, one of the few companies in this sector in Europe, that manucture MMW cameras are involved. These companies allows the project to capture, in its facilities, and in our laboratory at UGR, the images it needs. In the project also participates Prof. A.K. Katsaggelos, Director of Motorola Center for Seamless Communications en la Northwestern University, who provides PMMWI acquired by using Compressive Sensing techniques.

The project is based on the following assumptions. Processing passive millimeter images need to be addressed through the use of Bayesian techniques. These techniques are also required for passive millimeter processing images that have been captured using compressive sensing techniques to reduce acquisition time. Automatic extraction of semantic information, threats in our case, from these images can be approached through the use and adaptation of machine learning models on them. Although machine learning methods for extracting semantic information from passive millimeter images could be improved with new data or different training methods, the greatest improvement will necessarily associated with the interaction between the areas of processing and extraction of semantic information. In the project, for its interest, the type of semantic information is extracted threat detection although the proposed methodology is applicable to the extraction of other types of semantic information.

Therefore, the project aims, using our multidisciplinary group, achieving three general objectives involving the research and development of algorithms and methods:

• GO1: processing of passive millimeter images

• GO2: processing images captured with compressive sensing techniques

• GO3: threat detection

To achieve the overall objective GO1 processing passive millimeter images, research and development of methods, mainly Bayesian, blind deconvolution, superresolution and image fusion is proposed. Processing of PMMWI that have been captured using compressive sensing techniques (GO2) and which therefore have only a set of observations (projections) through a sampling matrix, research and development of methods of blind deconvolution methods and superresolution of these observed projections is proposed these observed projections.

For the purpose of detecting threats (GO3) in passive millimeter images, creating a large database of these images is proposed, the application and adaptation of machine learning methods and processing / detection interaction to improve detection system.

The project believes it can reach the following conclusions:

• Models and methods of Bayesian inference to improve image quality of PMMWIs

• The enhanced images facilitate visual inspection tasks thereof and the automatic extraction of semantic information

• The success rate of semantic extraction methods (detection of threats) we propose improve current methods based on the use of models of active contours.

• The processing / information retrieval interaction improves the success rate of the proposed methods.