Anti-Personal-Landmine Detection

Abstract - Over 64 countries worldwide are contaminated with land mines and there are in excess of 100 million land mines in the ground. Every day casualties and fatalities are being caused by land mines. In particular mines with a low metal content are difficult to detect since they can not be found using conventional metal detectors. A wide range of different approaches has been addressed including, Electro-optical approaches, NQR, GPR, and ultra wide band radar. Due to it's ground penetrating capabilities radar is a most promising technique for the detection of buried objects. However, a key challenge in remote mine detection is the discrimination between mines and non-mine objects (e.g. stones). The European Joint Research Center (JRC) in Ispra, Italy has performed a range of polarimetric wide-band radar measurement on mines and mine like objects. Based on these data sets the scattering mechanisms of mines and mine like objects are analyzed using various decomposition techniques in order to find differences in the scattering behavior to discriminate between mine and non-mine objects.

keywords: detection of buried objects, land mines, non mine targets

Polarimetric Features

From S-matrix we can 2 parameters can be derived yielding information for the identification of the scattering mechanisms. The first one is the $\alpha$ - angle,

$$\alpha=\arccos \left( \frac{S_{HH} + S_{VV}}{2 \sqrt{span(S)}}\right)$$ (5.2)

which is a continuous parameter with a range from zero to ninety degrees and can be used to represent a wide variety of different scattering mechanisms. Please note that the formulation of the alpha angle here is different from the formulation we derived in the decomposition part.

The second parameter is the vorticity $\nu$. The vorticity contributes to understanding of target structure in a way that is complementary to information in the coherent target vector. As pointed out in the decomposition part, the vorticity is calculated as

$$\nu = \frac{span(S)-\det(S)}{span(S)+\det(S)}$$

$$\mathrm{where} det(S) = \left\vert (S_{HH} * S_{VV}) - (S_{HV} * S_{VH})\right\vert$$ (5.3)

$$\mathrm{and} span(S) = \left\vert S_{HH}\right\vert^2+\left\vert S_{HV}\right\vert^2+\left\vert S_{VH}\right\vert^2+\left\vert S_{VV}\right\vert^2$$

Experimental Results

Due to the different nature of the measurements and the desired goal, some restrictions apply. For the detection and discrimination of buried objects only frequencies up to 4-5 GHz are useful, due to the limited penetration depth for higher frequencies. Therefore, we concentrated our analysis on the frequencies below 5 GHz. All data sets where analyzed, using various decomposition approaches. The most promising techniques were the alpha angle [Cloude96] and the vorticity concept [Bebington].

Anti Personal Mines (APL)

Since a real mine fields may contain APLs of different types which are not known a priori it is essential to find features which are common for all mines. Since, APLs vary in size shape and materials these properties are unlikely to be useful for mine detection. However, nearly all mines contain an air gap inside, which could be a key for remote mine detection. For APLs the JRC database provides measurements for 5 different APLs from 10 different azimuth [ $0^\circ - 360^\circ$] and 10 incidence angles [ $0^\circ - 90^\circ$], over a frequency range from 1GHz up to 10GHz. For oblique incidence angles the mines show at some higher frequencies high alpha and vorticity values. For close to vertical incidence angles the features disappear. Those scattering processes are unlikely to occur for surfaces [Nesti], and could therefore, be used to detect APLs. The features are independent of the azimuth angle. As shown in Fig. 5.6.

Non-Mine Targets

A main problem in remote mine detection is not only the detection of APLs, but also the discrimination between mines and non-mine targets in order to reduce the false alarm rate to a reasonable level without reducing the actual mine detection sensitivity. The JRC data base provides polarimetric radar measurements over a wide frequency range for a variety of non-mine targets which are likely to be found amongst APLs in a real mine field (e.g. Aluminum cans, plastic bottles, stones, wood etc.). Currently the Ultra-Wide-Band free space measurements are only available for a horizontal incidence angle. In order to discriminate between such targets and real mines the features for both types must different to allow a mine/non-mine decision with a sufficient accuracy to detect all mines as well as keeping a reasonable false alarm rate. The analysis of the mine like objects showed, the alpha and vorticity features differ substantially from those of the APLs. The most non-mine targets show also high values for alpha and vorticity but the distribution for these high values differs significantly from those of the APLs. The distribution of the higher values for alpha and vorticity is obviously a function of the azimuth angle. But also for specific azimuth angles the distribution of the high values of alpha and vorticity differs from those of the APLs. While the APLs show some clear narrow bands in the frequency range where high values occur, for the non-mine targets the high values are more spread out. Only for non-metallic objects with an air gap inside (i.e. plastic bottle shown in Fig. 5.7 (bottom)) the features show some similarities, since the structure of these objects (non-metallic cylinder with air gap inside) is similar to the structure of APLs.

However, the results below provide sufficient evidence to prove that discrimination between APLs and non-mine targets is possible by using polarimetric wide band radar measurements.

Surfaces

Since the JRC database currently does not provide wide band polarimetric radar measurements of surfaces at oblique incidence angles, we used an other set of surface measurements, also provided by JRC [Nesti]. The measurements show for all incidence angles clear surface like behavior. The higher values for one specific azimuth angle are most likely a measurement error due to reflections from the tank (missing absorber etc.) Even if the alpha angle spectrum widens for larger incidence angles as expected [Cloudemines] the values and their distribution are clearly different from those of the APLs and the non-mine targets as shown in Fig. 5.8. Such difference in the scattering behavior can - at least theoretically - be used to detect APLs and non-mine targets on top or buried beneath a surface.

Figure 5.6: Anti Personal Mine (APL) results for $\alpha$ - angle (left) and vorticity $\nu$ (right) over azimuth [ $0^\circ - 360^\circ$] (upper images) and incidence angle (vertical - horizontal[ $0^\circ - 90^\circ$]) (lower images) over a frequency range from 0.5 GHz (top) to 4.5 GHz (bottom of each image)

Figure 5.7: Non-Mine Target results for $\alpha$ - angle (left) and vorticity $\nu$ (right) for a stone (top) and a plastic bottle (bottom) over azimuth [ $0^\circ - 360^\circ$] over a frequency range from 0.5 GHz (top) to 4.5 GHz (bottom of each image)

Figure 5.8: Surface results for $\alpha$ - angle (left) and vorticity $\nu$ (right) smooth (top) and rough (bottom) surfaces over a frequency range from 0.5 GHz (top) to 4.5 GHz (bottom of each image)

Conclusions

The results discussed above indicate, that polarimetric wide band radar stand-off measurements can provide substantial information for the detection of APLs as well as for the discrimination between APLs and non-mine targets at oblique incidence angles. Since the features for APLs and surfaces differ significantly, a detection of surface laid mines is theoretically possible by means of polarimetric wide-band radar measurements. In order to validate these findings more measurements; i.e. on surface laid mines (and in a further step also of buried mines) are necessary. Also the scattering behavior and therefore, the features of the various non-mine targets differ recognizable. More measurements on such objects could enhance the understanding of the scattering mechanisms and could lead to a higher classification accuracy and therefore to a lower false alarm rate, or even yield the possibility do discriminate between different types of non-mine targets. The azimuth variation provides the most obvious and easiest discrimination between mine and non-mine targets, but is not likely to be suitable or at least very restricted (e.g. width of the vehicle) for real applications. However, this information might be useful for larger sensors like airships for the detection of mine fields.