Hyperspectral Imaging Tera Hertz System for Soil Analysis: Initial Results [original]

sensors

Article

Hyperspectral Imaging Tera Hertz System for Soil

Analysis: Initial Results

Volker Dworak 1,*, Benjamin Mahns 1, Jörn Selbeck 1, Robin Gebbers 1and

Cornelia Weltzien 1,2

1Department Engineering for Crop Production, Leibniz-Institute for Agricultural Engineering and

Bioeconomy (ATB), Max-Eyth-Allee 100, 14469 Potsdam, Germany; [email protected] (B.M.);

[email protected] (J.S.); [email protected] (R.G.); [email protected] (C.W.)

2Faculty V of Mechanical Engineering and Transport Systems, Chair of Agromechatronics,

Technische Universität Berlin, Strasse des 17. Juni 136, 10623 Berlin, Germany

*Correspondence: [email protected]; Tel.: +49-331-5699-420; Fax: +49-331-5699-849

Received: 7 August 2020; Accepted: 30 September 2020; Published: 3 October 2020





Abstract:

Analyzing soils using conventional methods is often time consuming and costly due to

their complexity. These methods require soil sampling (e.g., by augering), pretreatment of samples

(e.g., sieving, extraction), and wet chemical analysis in the laboratory. Researchers are seeking

alternative sensor-based methods that can provide immediate results with little or no excavation and

pretreatment of samples. Currently, visible and infrared spectroscopy, electrical resistivity, gamma ray

spectroscopy, and X-ray spectroscopy have been investigated extensively for their potential utility in

soil sensing. Little research has been conducted on the application of THz (Tera Hertz) spectroscopy

in soil science. The Tera Hertz band covers the frequency range between 100 GHz and 10 THz of

the electromagnetic spectrum. One important feature of THz radiation is its correspondence with

the particle size of the fine fraction of soil minerals (clay <2

m to sand <2 mm). The particle size

distribution is a fundamental soil property that governs soil water and nutrient content, among other

characteristics. The interaction of THz radiation with soil particles creates detectable Mie scattering,

which is the elastic scattering of electromagnetic waves by particles whose diameter corresponds

approximately to the wavelength of the radiation. However, single-spot Mie scattering spectra are

difficult to analyze and the understanding of interaction between THz radiation and soil material

requires basic research. To improve the interpretation of THz spectra, a hyperspectral imaging

system was developed. The addition of the spatial dimension to THz spectra helps to detect relevant

features. Additionally, multiple samples can be scanned in parallel and measured under identical

conditions, and the high number of data points within an image can improve the statistical accuracy.

Technical details of the newly designed hyperspectral imaging THz system working from 250 to

370 GHz are provided. Results from measurements of different soil samples and buried objects

in soil demonstrated its performance. The system achieved an optical resolution of about 2 mm.

The sensitivity of signal damping to the changes in particle size of 100

m is about 10 dB. Therefore,

particle size variations in the

m range should be detectable. In conclusion, automated hyperspectral

imaging reduced experimental effort and time consumption, and provided reliable results because

of the measurement of hundreds of sample positions in one run. At this stage, the proposed setup

cannot replace the current standard laboratory methods, but the present study represents the initial

step to develop a new automated method for soil analysis and imaging.

Keywords: hyperspectral imaging; Mie scattering; soil imaging; soil sensing

Sensors 2020,20, 5660; doi:10.3390/s20195660 www.mdpi.com/journal/sensors

Sensors 2020,20, 5660 2 of 22

1. Introduction

Soil is a fundamental resource in the earth’s ecosystems and for human life. The understanding

of the soil’s status and function is highly relevant because soils contribute to the recycling, filtering,

transformation, and buffering of substances, in addition to the production of food, forage, and biogenic

raw materials. In particular, for sustainable agriculture soils must be analyzed regularly to assess

their fertility. However, soil analysis with conventional sampling and laboratory-based methods

is expensive and time consuming. Thus, scientists are searching for new, sensor-based methods

that can analyze soils more efficiently. The development of soil sensors is a challenging task due to

the complexity of the soil and its many interfering parameters. This is also true for proximal soil

sensing [

]. To date, visible and infrared spectroscopy, electrical resistivity, gamma ray spectroscopy,

and X-ray spectroscopy have been investigated extensively [

]. One versatile candidate sensor for

performing nondestructive soil measurements is Tera Hertz (THz) spectroscopy, which covers the

electromagnetic frequency range between 100 GHz and 10 THz [

]. An important feature of THz

radiation is its correspondence with the particle size of the fine fraction of soil minerals (clay <2

to sand <2 mm). The particle size distribution is a fundamental soil property which governs soil

water and nutrient content, among other characteristics. Another important feature of THz radiation

is its ability to penetrate materials. Furthermore, it is nondestructive and not hazardous. However,

little research has been conducted on the application of THz spectroscopy in soil science. Moreover,

the development of THz systems is ongoing. Some non-continuous wave systems are used and

the applications are in their infancy [

]. Even fewer studies have been performed on continuous

wave (CW) THz systems [

] and hyperspectral imaging [

] in soil samples. The first results using

THz radiation for soil analysis in the range of 258–375 GHz demonstrated a specific interaction with

soil particles [

]. In this frequency band, the particle size in the range of a millimeter causes Mie

scattering [

–

]. Mie scattering, named after the physicist Gustav Mie, is the elastic scattering of

electromagnetic waves by particles whose diameter corresponds approximately to the wavelength

of the radiation. In the case of multiparticle scattering, Mie scattering shows a complex spectral

behavior and varies depending on the local position of the beam. Hyperspectral imaging overcomes

the problem of the local variability within the sample and the problem of the interpretation of a single

measurement because it combines imagery and spectral behavior. Therefore, most hyperspectral

applications and research result from this combination [

–

]. The human eye can easily interpret the

image and identify different areas, interface regions, and buried objects. Therefore, THz imaging is

used for defect or artifact detection [

]. Additionally, critical samples can be simultaneously measured

under identical conditions, with fewer time effects caused by the long measuring time. Conversely,

the time effects can be compared in a simultaneous measurement, and exchanges between different

samples can be analyzed. The measurement of multiple data points helps to overcome individual

effects, and common areas can be interpreted statistically. In this way, new images can be created.

A disadvantage of this imaging method is the additional complexity of the focus plane. Sharp results

can be produced only for features in the focus plane. The high Mie scattering of sand particles prevents

beam formation in the material, and a focus plane does not exist under such conditions. This makes

generating imagery of soil samples difficult. In order to develop a THz measurement system for a

challenging task such as soil characterization, an experimental setup must be developed that allows

differentiating and isolating the many different influencing parameters, and at the same time enables an

imaging approach by precise localization of each measurement reading within the sample’s dimension.

Such an experimental setup will enable basic research on the potential of soil sample characterization

through hyperspectral analysis of THz radiation transmission/reflection patterns. This paper describes

the development of a THz hyperspectral imaging experimental system for soil sample characterization.

With this setup, the authors note the advantages of this method for testing complex samples such as

soil samples. This paper discusses the following research questions:

•Can the effect of scattering on image quality be demonstrated by hyperspectral imaging?

Sensors 2020,20, 5660 3 of 22

•Can the imaging localize artifacts or measurement errors?

•How can the imaging identify homogenous sample regions for statistical comparisons?

2. Materials and Methods

Hyperspectral imaging of difficult samples with terahertz radiation is a complex task because of

the influence of many physical parameters. It starts with the complicated behavior of the Rayleigh

and Mie scattering of thousands of particles. In previous studies, the CELES software [

] enabled

the simulation of such complex Mie scattering arrangements and can be used to visualize the results.

However, the simulation is not a part of the current work; rather, the measurement possibilities of the

presented setup will be demonstrated. The experimental setup described consists of the soil samples

and holders, the THz spectrometer, the sample positioning system, the operating software, and the

data analysis.

Additionally, natural soil is a complex sample material and soil functionality depends on multiple

parameters, including particle size distribution, mineral content, carbon content or organic matter,

water, biology, and physical and chemical properties. Therefore, all first measurements were made on

simplified soil samples to reduce the number of parameters. The results in this article focus on the

imaging possibility of hyperspectral THz measurement.

2.1. Soil Samples and Holders

The measurement is influenced by the sample holder, the preparation of the sample, the filling

of the sample holder, and the compaction (process) in the sample holder. Even the measurement of

the empty sample holder is not adequate for calibration because of the different interface situation

at the sample holder walls. The best filling substance for calibration is oil [

], but this discussion is

not part of this work. The sample holders demonstrated may not be the best solution but rather are

a good practical approach. All samples in this study were placed in sample holders made of HDPE

(high density polyethylene) with a wall thickness of 2 mm. The holders were parallel, box-shaped,

or wedge-shaped. The wedge-shaped sample holder enables simultaneous measurements of different

thickness. The sample thickness was addressed accordingly. In some cases, the sample holder was

separated into domains with different samples (Figures 1–4) to demonstrate the resulting contrast in

the corresponding image. Figure 1shows a box-shaped sample holder with a 10 mm sample thickness.

The box was divided with a 2 mm thick piece of paper and filled with quartz particles in different

size fractions.

Sensors 2019, 19, x FOR PEER REVIEW 3 of 23

2. Materials and Methods

Hyperspectral imaging of difficult samples with terahertz radiation is a complex task because of

the influence of many physical parameters. It starts with the complicated behavior of the Rayleigh

and Mie scattering of thousands of particles. In previous studies, the CELES software [25] enabled

the simulation of such complex Mie scattering arrangements and can be used to visualize the results.

However, the simulation is not a part of the current work; rather, the measurement possibilities of

the presented setup will be demonstrated. The experimental setup described consists of the soil

samples and holders, the THz spectrometer, the sample positioning system, the operating software,

and the data analysis.

Additionally, natural soil is a complex sample material and soil functionality depends on

multiple parameters, including particle size distribution, mineral content, carbon content or organic

matter, water, biology, and physical and chemical properties. Therefore, all first measurements were

made on simplified soil samples to reduce the number of parameters. The results in this article focus

on the imaging possibility of hyperspectral THz measurement.

2.1. Soil Samples and Holders

The measurement is influenced by the sample holder, the preparation of the sample, the filling

of the sample holder, and the compaction (process) in the sample holder. Even the measurement of

the empty sample holder is not adequate for calibration because of the different interface situation at

the sample holder walls. The best filling substance for calibration is oil [26], but this discussion is not

part of this work. The sample holders demonstrated may not be the best solution but rather are a

good practical approach. All samples in this study were placed in sample holders made of HDPE (high

density polyethylene) with a wall thickness of 2 mm. The holders were parallel, box-shaped, or wedge-

shaped. The wedge-shaped sample holder enables simultaneous measurements of different thickness.

The sample thickness was addressed accordingly. In some cases, the sample holder was separated into

domains with different samples (Figures 1–4) to demonstrate the resulting contrast in the

corresponding image. Figure 1 shows a box-shaped sample holder with a 10 mm sample thickness. The

box was divided with a 2 mm thick piece of paper and filled with quartz particles in different size

fractions.

Figure 1. HDPE (high density polyethylene) sample holder for 10 mm sample thickness. A two

millimeter thick piece of paper separates the holder in the middle. The left chamber is filled with the

63–100 µm quartz fraction, and the right chamber is filled with 400–500 µm quartz particles.

The free configuration of the scan positions helps to reduce measurement times. For a sample

preparation as shown in Figure 1, only the interface regions are sampled with smaller step widths for

higher resolution. Areas in a homogenous sample region can be represented with a few measurements

and can be scanned with a larger step width.

Figure 1.

HDPE (high density polyethylene) sample holder for 10 mm sample thickness.

A two millimeter thick piece of paper separates the holder in the middle. The left chamber is filled with

the 63–100 µm quartz fraction, and the right chamber is filled with 400–500 µm quartz particles.

Sensors 2020,20, 5660 4 of 22

The free configuration of the scan positions helps to reduce measurement times. For a sample

preparation as shown in Figure 1, only the interface regions are sampled with smaller step widths for

higher resolution. Areas in a homogenous sample region can be represented with a few measurements

and can be scanned with a larger step width.

Sensors 2019, 19, x FOR PEER REVIEW 4 of 23

Additionally, the dependency of the sample thickness can be analyzed simultaneously if a wedge-

shaped sample holder is used (Figure 2). The sample holder was filled with quartz sand of three

different particle sizes. The overlay in Figure 2 with the blue lines indicates the boundary between the

three samples. These sample holders have a length-to-thickness ratio of 5:1. The smallest sample holder

started at zero thickness, the middle sample holder started at 5 mm, and the largest sample holder

started at 10 mm.

Figure 2. Wedge-shaped sample holder filled with three different soil samples. The blue horizontal

lines show the lines between the samples. Samples are quartz sand with different particle sizes. At

the bottom is the 100–200 µm fraction, in the middle is the 400–500 µm fraction, and on top is the 500–

600 µm fraction. The red rectangle shows the measurement area for this example.

Additionally, the sample holder can be filled with different materials, as shown in Figures 3 and

4. Therefore, identical measurement conditions are established, and quantitative differentiation is

enabled. For samples that change over time, the fastest scan line across the different materials is selected

to minimize the time difference. Every scan pattern for the image can be implemented by the user (see

Section 2.4).

Figure 3. HDPE sample holder for 10 mm sample thickness filled with four different soil samples. The

topsoil is pure Luvos

Healing Earth. The next sample is Luvos with 20% sulfur added. The third

Luvos

P2O5

K2CO3

Figure 2.

Wedge-shaped sample holder filled with three different soil samples. The blue horizontal

lines show the lines between the samples. Samples are quartz sand with different particle sizes. At the

bottom is the 100–200

m fraction, in the middle is the 400–500

m fraction, and on top is the 500–600

fraction. The red rectangle shows the measurement area for this example.