The rising demand for food crops and the growing concern for environment have made it necessary to shift from traditional agricultural practices towards modern agricultural practices. Traditional agricultural practices are labor intensive, time consuming, expensive and also a cause of environment pollution. To achieve sustainable agriculture, it is necessary that the precision agriculture technologies and practices are integrated with the traditional practices, which will also help to deal with the spatial heterogeneity of the soil [1]. The biggest hurdle in the proper implementation of precision agriculture is the inability to generate information related to a particular site rapidly and at an acceptable cost using laboratory analysis and soil sampling methods.

The nutrients required for the healthy growth of a crop are obtained from the soil. The quality of crop yield depends on the quality of soil in which it grows. Therefore, soil testing is an important aspect of precision agriculture. The proposed research work is an effort towards the design and development of a soil monitoring system that can be used to estimate the urea content in soil. The system makes use of Partial Least Squares Regression Technique (PLSR) for the estimation of urea. The system can be made portable, smart, low cost and user friendly through the use of embedded systems. With some modifications the system can be designed to estimate more than one soil component. The thesis is organized in the following chapters as described below.

Chapter I (Introduction)

“Agriculture not only gives riches to a nation, but the only riches she can call her own”[2]. The growth in the demand for food, feed and fiber globally is anticipated to grow by 70 percent. The demand for crops for industrial use and in the production of bio-energy is also expected to rise simultaneously. The increasing demand for agricultural goods will put huge pressure on the limited resources available. The increase in urban settlement areas will force agriculture to compete for land and water. Agriculture will therefore have to adapt itself to newer conditions and at the same time will have to address issues related to climate change, maintenance of biodiversity and preservation of natural habitats [3]. To meet these demands, farmers therefore need to equip themselves with new technologies so as to increase productivity with limited number of resources.

Sustainable resource management is the need of the hour. Conservation of soil quality is crucial to sustainability in agriculture. This has led to a shift from the use of traditional agricultural practices to modern agricultural practices so that the available resources are utilized in a sustainable manner. The modern technique of farming known as precision farming is based on the concept of site specific crop management. This method takes into consideration variability exhibited by the soil and accordingly inputs are applied based on the local requirements within a field. Soil sensing plays an important role in precision farming. Large numbers of soil sensors are being developed all around the world to measure different soil properties. Some of which are still in the research and development stage and some of which are commercially available.

Based on their principle of working these soil sensors can be classified as follows:

• Electrical and Electromagnetic sensors: Depending on the composition of soil under test, electrical capacitance or inductance, resistivity or conductivity of the soil is measured. The response time of these sensors is very fast, they have high durability and are of low cost. These sensors are commercially available.
• Optical and Radiometric sensors: These sensors, through the use of electromagnetic waves, measure the level of energy that is either absorbed or reflected by the soil particles depending on the soil composition. The properties of the soil are measured using visible and near-infrared wavelengths [4]. They can be used for the estimation of CEC, soil texture, moisture and other soil parameters with the help of appropriate data analysis techniques.
• Mechanical sensors: these sensors measure soil resistance with the help of a tool used in the soil. The measure of resistance offered by the soil has a relation with the compaction of the soil which is a spatially varying property of soil.
• Acoustic sensors and Pneumatic sensors: Though these are a class of mechanical sensors, they can be used as an alternative means for the differentiation of physical and mechanical characteristics of soil. Measurements taken using pneumatic and acoustic sensor have been used to correlate soil texture and compaction. The application of acoustic sensors in characterizing the physical state of soil is not very clear and requires more research work.
• Electrochemical sensors: These sensors produce an output voltage through the use of ion selective membranes, depending on the activity of ions under study such as H+, K+, NO3 −, Na+, etc. Soil pH sensors using this technique are already commercially available. The extraction of ions such as potassium in real time is still not possible even though the concept appears to be simple.

There is a need to develop fast, real time and portable soil sensors which can generate soil report instantly. Thus, the problem of designing and developing a smart soil monitoring system was formulated using a reconfigurable embedded system platform.

Chapter II (Literature Survey and Objectives)

The conventional laboratory methods of soil testing have a number of limitations such as they are expensive, labor intensive and time consuming. As such new methods of soil testing are being developed across the globe. A number of soil nutrient sensing techniques are in the stage of development and testing. These sensors can be broadly classified into two types depending on the techniques of measurement being used.

1. Optical sensing uses reflectance spectroscopy technique wherein the light that is absorbed/reflected by soil particles is measured. Since optical sensing techniques have the advantage of being non-destructive they are more widely used as compared to electrochemical sensing techniques [5], [6]. Soil color analysis can be used for estimating soil organic matter content through the use of optical sensors [7]. The visual and near-infrared spectral reflectance in optical sensing can be used for estimating soil texture, moisture, CEC etc. [8].

2. Electrochemical sensing is based on the measurement of current or voltage generated between the sensing electrode and the reference electrode. The amount of voltage or current measured is related to the concentration of the selected ions such as H+, K+, NO3-, etc. [8]. Ion selective electrodes made of glass or polymer membrane, or ion-selective field effect transistors are used for the measurement of soil fertility. Ion-selective membrane sensors have a huge potential in the development of on-the-go soil nutrient(s) and pH sensors [9]. Currently, the accuracy of the results using these sensors is low as compared to those using laboratory tests, but this can be taken care of by increasing the sampling density.

Use of Spectroscopic techniques in the estimation of soil properties has been demonstrated since 1970’s [10]. Various methods using spectral analysis have been proposed for the measurement of the soil properties. Methods that are based on the physical and analytical characteristics of the signal and chemometric based empirical methods provide good effective predictability. Therefore, the relation between soil properties and soil absorption can be used to develop regressions using field and laboratory data for calibration. Spectroscopic techniques are found to be faster, can provide real time measurements and are of low cost, as compared to conventional methods and hence are found to be more suitable when there are more samples and analysis to be done. Also, unlike laboratory testing methods which require sample pre-processing and the use of chemical extractants, spectroscopic techniques can be used directly, thus saving on cost and time [11].

Thus, the problem of developing a soil nutrient sensor using RF spectroscopy based on the dielectric principle was formulated. The thesis emphasizes on the design and development of the sensor and the use of embedded platform to make it portable, real time and user friendly system through the use of DSP algorithms.

Objectives:

In order to meet the global requirements of increased crop productivity and sustainable agriculture, there is an urgent need of developing soil sensors which are fast, accurate and portable. Also, the problem was formulated keeping in mind the conditions of Indian farmers. Indian farmers are mainly small farmers who are poor, technically unfit and cannot afford modern tools. This research work is being undertaken with the main objective of developing a fast, portable, cost effective and user friendly soil monitoring system to analyze the fertility status of the soil.

The objectives of the research work are the design of a dielectric cell to measure absorption loss at RF frequencies for various soil nutrients and to use this RF data to develop a FPGA based smart soil monitoring system for accurate prediction of soil content using PLSR technique. The system shall be user friendly as well as reprogrammable for changed environmental conditions.

Chapter III (System Design for Soil Monitoring System)

The block diagram of proposed design for Soil Monitoring System is as shown in Figure 1. The design consists of RF data obtained from Scalar Network Analyzer fed as input to Altera DE2 board with target as NIOS II FPGA. The RF data is obtained from the soil sensor connected between a tracking generator and a spectrum analyzer.

A soil sensor based on the dielectric loss technique is designed and constructed to measure the RF responses of various soil nutrients. The cell is rectangular in shape with outer dimensions 13cmx2cmx2.5cm and is made up of PMMA sheets. The inside surface of the cell is lined with gold foil and the same is connected to the outer shield of the feed connectors so as to provide the necessary shielding effect. The outer surface of the cell is covered with a copper foil and is also provided with the necessary shielding effect. A wire made of gold is connected from the input feed connectors to the output feed connector at centre of the cell. The RF spectrum of a sample is measured by placing it in the cell. A tracking generator is used for injecting an RF signal into the sample through the central gold wire. Thus, a dielectric cell consisting of the central wire, the outer copper shield and the sample is formed. The signal strength starts reducing as it propagates through the central wire from the input end to the output end of the cell, due to the dielectric loss associated with the sample solution. Thus, an output signal proportional to the absorption loss of the sample solution is captured by the RF spectrum analyzer connected at the receiver end of the cell. Signal Hound USB-TG44A tracking generator and Signal Hound USB-SA44B spectrum analyzer are used with both the instruments working in the frequency range of 1Hz-4.4GHz.

Figure 1: Block diagram of the Soil Monitoring System

Figure 2 shows the RF spectra for urea in the range 10MHz to 4.4GHz.

Figure 2: RF Spectra of urea.

Figure 3: Section of urea spectra with varying concentrations

Samples for obtaining the RF responses of various soil components are prepared by dissolving the required component in distilled water. The amount of the component to be added to water was calculated from the data obtained from agricultural department. This amount was taken as the normal concentration of a particular component found in the soil. Samples of varying concentrations of the soil components are prepared and denoted as 1 for normal, 0.5 for half the normal, 2 for twice and 3 for thrice. The soil components considered for the study are urea, potash, phosphate, calcium carbonate and sodium chloride.

The frequency range of 10 MHz-4.4GHz is divided into smaller frequency ranges based on the unique frquencies at which the variation in the attenuation is found as per the change in the concentration of the soil component. A set of recorded spectra for various combinations of the five soil components with concentrations ranging from 0.5 to 1.5 are used in the calibration file. In order to predict the unknown concentration of urea in a sample, the detected spectra containing the urea signature along with the other components is passed through signal conditioning stage. The output from Spectrum Analyzer is stored in the computer. This data is then fed to a CYCLONE II device with Altera Nios II processor running on it. The recorded spectra are then passed through SIMPLS algorithm running on NIOS II processor. The algorithm predicts the concentration of unknown sample (Urea) and displays the result on LCD or a computer screen. The SNR of the detected spectra must be sufficiently high so as to provide reliable urea specific information and therefore data processing is needed to identify spectral features of urea from the combination spectra originating from interfering matrix components like potash, phosphate, sodium chloride and calcium carbonate.

We can extend the use of this system for the analysis of other soil components by modifying the processing algorithms required to analyze that component without changing the hardware.

Chapter IV (Multivariate Data Analysis)

It is a statistical analysis technique used in the case of data consisting of multiple variables. Due to the advancements in the field of information technology there is a huge amount of data being generated in various fields. Though the magnitude of data available is huge, it is still a challenge to derive useful information and knowledge from this data.

Multivariate Analysis can be used to derive meaningful information for the improvement of process performance and product quality. Over the last decade, multivariate analysis is being successfully used to monitor and model chemical/biological processes [12].

Techniques using multivariate data analysis are widely used in the analysis of spectral data both quantitatively and qualitatively. Quick analysis of complex samples from their spectral signatures can be done using standard tools like Partial Least Squares (PLS), Principal Component Regression (PCR), Principal Component Analysis (PCA), Multivariate Curve Resolution (MCR) and discriminant analysis based on chemometric techniques [13]. Partial least squares (PLS) isone of the recent multivariate data analysis technique particularly useful in situations where there is a large set of independent variables (i.e., predictors). A set of dependent variables can be predicted from this set of independent variables by using PLS. Partial Least Squares (PLS) can be an effective tool for the analysis of data as it has minimum constraints on scales of measurement, size of sample, and residual distributions. It consists of methods for regression and classification, and techniques for reducing dimension and tools for modeling. The basic assumption on which the PLS methods work is that a small number of latent variables that are not directly observed or measured are used to drive the observed data from a process or a system. The technique of PLS for projecting the observed data to its latent structure was developed by Herman Wold and coworkers. PLS is now being used as a standard tool in the analysis of a wide spectrum of chemical data problems in chemometrics. The successful data analysis of PLS in chemometrics has led to its increase use in other scientific fields such as bioinformatics, food research, medicine, pharmacology, social sciences, physiology etc.

PLS is a multivariate technique that transforms the input-output data onto a smaller latent space, by extracting a small number of principal factors having an orthogonal structure. The extraction of the factors is done in such a way that it provides maximum correlation with the dependent variable [14]. To model linear relations between multivariate measurements, PLS is used as a standard tool.

Multivariate Calibration Model for Soil Monitoring System:

Multivariate spectroscopic data can be analyzed using the PLSR model. PLSR is one of the techniques of multiple linear regressions and is probably the least restrictive of the various multivariate techniques used in multiple linear regression models. This feature of PLSR makes it possible to be used in situations when there are limitations on the use of other multivariate methods, for example, when the predictor variables are many as compared to number of observations.PLSR can be used as an elementary analysis tool for the selection of suitable predictor variables and in the identification of outliers. PLSR model based on SIMPLS algorithm using C language is developed and ported on NIOS II platform to estimate the urea concentration.

The PLSR model is validated for the following cases:

Case 1: Changing urea concentrations from below normal to above normal i.e. from 0.5 to 2 and keeping other components at their normal concentration value i.e. 1.

Case 2: Changing the concentration of each of the other soil component from 0.5 to 2 and keeping urea constant in all the cases.

Chapter V (Design of FPGA Soft Cores for Soil Monitoring System)

DSP functions can be implemented using two different types of programming platforms: digital signal processors (DSP) and field programmable gate arrays (FPGAs). Digital signal processors are microprocessors specifically designed for handling DSP tasks, while FPGAs are reconfigurable signal processors.

The factors that make FPGAs more suitable, particularly for high performance computing applications are: (i) Huge potential for implementation of parallelism (ii) The control logic is embedded (iii) On-board memory in FPGA helps to overcome the limitation set by number of I/O pins on processor logic’s memory access bandwidth and hence results into significant performance benefits (iv) A higher capacity FPGA can be used on the same board as an older device, to support performance upgrades.

DSP Implementation on Embedded system

The implementation of DSP algorithms is done on Altera platform. A Nios II system is designed to measure the concentration of urea in soil. The Nios II system is the heart of the instrument which controls the various modules of the system like interacting with 12 bit ADC and performing the SIMPLS algorithms on the spectral data to estimate the concentration of urea. The whole interface and algorithms are implemented using 32-bit NIOS II soft-core ported on CYCLONE II FPGA.

Chapter VI (Analysis, Results and Conclusion)

The thesis covers the design and development of soil sensor based on the dielectric technique. The technique proposed the use of RF signals in the range of 10MHz-4.4GHz and analyzing the detected spectra in the soil sample for urea signature. In this thesis a novel Soil Monitoring System is developed using RF spectroscopy based on embedded technology. An Altera DE2 board based on NIOS II soft-core platform and having target as CYCLONE II (EP2C6) is used to estimate the urea content in soil in the RF range of 10MHz-4.4GHz. SIMPLS algorithm for PLSR model is developed using C language and embedded on the NIOS II platform for the estimation of urea concentration. The designed sensor was tested for its precision by recording the spectra of a particular component over a number of times. The PLSR model was validated by calculating percentage error under various conditions. It was found that the predicted urea values showed percentage error which was within the acceptable levels required for device development.