Diffuse Reflectance Illumination Module Improvements in Near-Infrared Spectrometer for Heterogeneous Sample Analysis

This paper presents a portable and affordable prototype using a Digital Micro-mirror Device (DMD) based Near-Infrared Spectrometer and an improved diffuse reflectance illumination module (DRIM). The improved DRIM produced optical geometry parameters such as 3.5mm standoff distance (SD), 2mm depth of overlap illumination area (DOIR), and 4mm sample active illumination area (SAIA). It enables the single and multi-point scans to determine the crude content of various food quality parameters and homogeneity by averaging spatial inhomogeneities of raw material and heterogeneous sample mixtures placed at a standoff distance. The prototype outperformed the current portable NIRS by a factor of 2-3 in terms of optical throughput, signal-to-noise ratio, and baseline. The prototype’s repeatability was determined by assessing pure samples such as chalk powder, red chili powder, wheat, and groundnuts using scattering correction techniques and was computed <1% relative standard deviation (RSD). Partial least square regression (PLSR) was used to build a prediction model using around 100 randomly selected poultry feed samples with 10-20% moisture ranges-. Results of the experiments indicated values for the coefficient of determination as high as 0.991, and root mean square error was 0.32%, and a prediction accuracy with maximum deviation of <1%. The results indicated that the prototype was able to efficiently predict heterogeneous mixtures and food grains, provide new specifications for single and multi-point scan measurements, and this carries a lot of potential as a stand-alone or in-line food monitoring tool.


I. INTRODUCTION
In today's food industry, there is a requirement for a rapid, reliable, accurate, compact, and cost-effective mechanism for analysing food products using modern technology. The near-infrared spectrometer (NIRS) is a versatile analytical tool that combines chemometric techniques with the benefits of being simple, rapid, non-invasive, and nondestructive [1,2,3]. The non-destructive feature of NIRS enables food sample analysis without any prior processing [4,5,6]. To extract spectral information from the test sample, NIRS uses standard measurement modes that allow non-invasive radiation interaction with the sample. The measurement modes are determined by the sample form or position, and the sensing wavelength region. The most common traditional measurement modes are transmittance, transreflectance, and diffuse-reflectance [7,8]. Numerous studies have been conducted on new techniques, instrumentations, and methods to increase the ability of near-infrared (NIR) analysis to measure quality and safety parameters in a wide range of food products and processes [9,10,11,12]. For high-performing NIRS, improving illumination module geometries is a critical parameter [13,14]. Experimental studies on new techniques ultimately help turn laboratory equipment into portable commercial sample estimation equipment for static and in-line analysis. As a result of these research studies, large and small-scale industries, retailers, and academia may use the latest technology.
Most NIRS instruments with a traditional grating spectrometer have a fixed grating with a linear array detector (InGaAs) or a scanning grating with a single-element detector (InGaAs). Conventional linear image sensor spectrometers are costly, have low resolution, and often need a single or multiple-stage thermoelectric cooler (TEC). Simultaneously, the mechanical structure of a scanning grating-based spectrometer with a single element detector is complex, mainly where portability is concerned [15]. Whereas a multi-point NIRS, which uses fiber probes to provide spectral and spatial information in real-time has a lot of potential in the food industry [16,17]. A multi-point scan analysis, rather than a single-point analysis, is needed to overcome significant interference caused by water and physical attributes such as sample size, shape, and hardness [18]. A multi-point NIRS device based on a Fabry-Perot interferometer could be used in-line on dairy products for simultaneous monitoring and quality evaluation [19]. In the pharmaceutical industry, inhomogeneities in the process chamber of the hot melt coating process can be detected by placing fibers at different positions that enable multi-point NIRS measurements [20,21]. The methods listed above can accommodate multiple fiber probes and NIR spectrometers in real-time multi-point analysis, but they are complex and expensive to use. A new digital micro-mirror device (DMD)based NIRscan Nano Evaluation module (EVM) proposed by Texas Instruments (TI) is a viable alternative to grating-based NIRS and FT-NIRS [22] because of its high signal-to-noise ratio (SNR), wavelength selection ability, speed, mechanical stability, and low cost. The NIRscan Nano EVM is another popular and potentially useful NIRS analytical tool that can be used in single-point scan analysis in real-time when in direct contact with the test sample, but it is not suitable for single and multi-point scan analysis for a standoff distant sample. It also has other design limitations, such as depth of overlap illumination region (DOIR) and sample active illumination area (SAIA). If the discussed problems can be addressed, the performance of NIRS can significantly improve.
In this proposed work, we have overcome the challenges of TI's NIRscan Nano EVM. The first challenge is the difficulty in collecting high-quality spectral data from samples at a standoff distance. Because of the limitations of this instrument's illumination module, collecting highquality spectral data from samples at a standoff distance is difficult. That is, it facilitates single-point analysis and allows for accurate measurements for a sample within 0.75 mm against a sapphire window but is inadequate to accommodate sample holder and sample rotation mechanisms for multi-point scan and standoff distant sample studies when the sample is placed at a distance of more than 0.75 mm from the sapphire window [23]. We propose modifying the illumination module's optical geometry to the desired standoff distance to address this problem. To optimize the illumination module and its instrumental functionality, we have conducted research studies on SAIA, DOIR, and collection of diffuse reflected light from target samples positioned at standoff distances. The result of our studies is an improved portable ELICO NIRS prototype which allows for single and multi-point scans to average out spatial inhomogeneities of the samples like bulk and heterogeneous samples [24]. We evaluated the efficiency and efficacy of ELICO NIRS prototype to measure moisture content in feed formulation using experimental analysis and by making comparisons with existing TI's instrument and the ELICO NIRS prototype.

II. LIMITATIONS AND SCOPE OF IMPROVEMENTS
The NIR instrument design was influenced by the target sample, measurement mode, wavelength region selection, techniques, and application field [25]. To maximize the instrument's effectiveness, an NIR range of 900-1700 nm was selected. For sample content evaluation, premathematical and chemometrics were used to resolve the NIR region's disadvantages, such as low structural selectivity and sample composition determination from weaker absorption overtones.
The NIRscan Nano EVM is an analytical tool that primarily consists of: • Illumination module -used to illuminate the sample and capture diffuse reflection light.
• A polychromator with a fixed grating and a DMD that allows for wavelength selection.
• Collimated and focus opticsto guide diffused light collection.
• A single InGaAs Photodetector is used to detect light photons and convert them into electrical signals.
For the sample analysis, a diffuse reflectance illumination module in NIRscan Nano EVM was used. In this configuration, the sample was kept directly against the sapphire window. A light beam from two lens end lamps was focused on the sample, scattered within the sample, and returned to the surface after being absorbed by the sample, which is referred to as diffuse reflectance. Optics captures the diffuse reflectance and directs it toward the detector. The geometry of this configuration reduces specular reflection, which contains no chemical information. The light source and detector are placed on the same side of the sample, illuminating it at a 40-degree angle to avoid specular reflection collection. The light beam from two lens-end tungsten lamps is focused about 3mm away from the lamps, intersects at about 0.75 mm past the sapphire window, and creates a 2.5 mm diameter illumination spot where the sample is placed. The collection optics [26] capture diffused reflectance light from a 2.5 mm diameter illumination spot at the sample window. Fig. 1 shows the NIRscan Nano EVM's illumination module.

FIGURE 1. Schematic of the Illumination Module of the NIRscan Nano EVM
The existing illumination module has the following drawbacks: • If the sample is moved farther away from the window, the sample cannot obtain enough illumination for the device to conduct an effective scan.
• Laboratory analysis of grains and powder samples necessitates a sample holder with a thickness of around 1.5mm, producing inaccurate results.
• It enables single-point analysis of a small portion of the sample, which is difficult to do when analyzing a large number of inhomogeneous or heterogeneous products.
The above points indicate a scope to develop a new illumination module for analyzing a standoff distant sample and attempting multi-point scan analysis by scanning different sample areas consecutively to average spatial inhomogeneities.

A. INSTRUMENT DESIGN
Current investigations have identified limitations in existing equipment to fulfill Food industry requirements for multipoint scan analysis of standoff distant samples. Therefore, we developed a new illumination module and integrated it with the NIRscan Nano EVM's polychromator for this prototype. This illumination module was made with two tungsten lensend lamps with a double filament and collecting optics. A dual filament provides enough illumination for the distant sample while reducing the number of lamps needed for module downsizing. A front-end lens guides more light from the filament to the sample target area. Appropriate optics was used to capture more diffused light from the sample's targeted active area and direct it to the polychromator. The SD of the sample, SAIA, and DOIR is determined mainly by the lamp divergence angle, the angle between the lamp axis and the sample plane, and the distance between two lamps in the geometry.
Multiple configurations were developed for the targeted standoff distance by using a set of equations referred to "1to5". The optimized illumination module geometry was determined by comparing numerous designs, as shown in Fig. 2. The proposed modules and the NIRscan Nano EVM were simulated in Zemax using non-sequential ray-tracing method. In order to determine the DOIR and sample location for attaining maximum power, the illumination power at various standoff distances from the sapphire window was examined. Finally, the functional parameters of the proposed prototype using indigenized application software were evaluated.

FIGURE 2. Illumination Module Geometry
Where X1, X2, X3, and Y1 are: Overlap area at targeted sample plane =2( 1 + 2 ) (5) A larger standoff distance model requires a larger sample window, larger module size, and high-powered lamps. High-powered lamps consume a lot of power, making the development of battery-powered devices a challenge. Increasing the sample window and device size allows for more stray light and increases the price. Factoring all these challenges in this analysis, we propose an optimized geometry to meet current needs without sacrificing efficiency.

B. INSTRUMENT PERFORMANCE VALIDATION
Before allowing the instrument to establish a prediction model for sample analysis, it must first go through a series of tests to determine its functional parameters such as throughput (detectivity), SNR, baseline, repeatability, linearity, and accuracy. To compare and examine the impact of different geometrical illumination models on the results, throughput, SNR, and baseline measurements were done for both ELICO NIRS prototype and the NIRscan Nano EVM. The scan data in 'CSV' format allows the user to upload and download the spectra. Reference data was captured with a known diffuse reflectance standard (ref no. SRS-99, Labsphere) close to 99% reflection in the entire desired wavelength region.

1) THROUGHPUT (INTENSITY)
The throughput of both the ELICO prototype and the NIRscan Nano EVM was measured by capturing the NIR spectrum from a known standard diffuse reflectance standard.

2) SNR
Signal-to-noise ratio (SNR or S/N) is a measure that compares the level of the desired signal to the level of background noise. NIR spectrometer requires large SNR and is achieved by the collection of maximum light intensity. Large SNR allows detecting small changes on a large signal and provides more precise measurements. On both ELICO NIRS prototype and NIRscan Nano EVM, SNR is measured by recording 100 scans at 15 ms, 30 ms, and 60 ms integration times.

3) BASELINE
The initial baseline of the instrument is established by considering the effects of the environment and background noise. To get an accurate absorption profile, subtract the resultant baseline from the sample raw data. Next, a baseline test scan is performed at various gain settings across the entire wavelength range and compared for both instruments. Finally, the corrected signal for the peak-topeak fluctuations is determined.

4) REPEATABILITY
The ELICO NIRS prototype's repeatability is studied using inter-day and intra-day variations. First, repeatability is measured by analyzing two different particle sizes of pure powder samples, such as chalk and red chili, and two different grain samples, such as wheat and groundnut. Then, all samples are analysed at different timing during one day for intra-day variations and the same procedure is followed for two different days to study the inter-day variations.
Data pre-processing is essential in analyzing the samples with different particle sizes and building a calibration model in NIR spectroscopy analysis. With well-designed pre-processing techniques such as scatter corrections and derivative techniques, the instrument's performance is greatly improved. To analyze the sample data for repeatability and linearity tests, we need to apply scatter correction techniques such as multiplicative scatter correction (MSC) and Standard Normal Variate (SNV) technique to minimise the scatter effects originating from path length differences and particle size.
For measuring repeatability with and without data preprocessing techniques, the percentage relative standard deviation (% RSD) of the concentration of 10 repeated scans of each sample has been taken in the multi-point scan method.

5) LINEARITY & ACCURACY
For studying the linearity and prediction accuracy of ELICO NIRS prototype, poultry feed samples were taken to analyze moisture content in feed formulation and the experiment collected around 100 samples from a feed manufacturing plant. On one set of the samples, we performed wet chemical analysis. On the second set, spectral measurements were taken using ELICO NIRS spectrometer having a range of 900 to 1700 nm. After taking the initial raw spectral measurements, the sample was spiked by adding moisture (water), then another spectral measurement was taken to obtain and validate the variation in moisture. After collecting raw data, preprocessing techniques were applied, and Partial Least Square Regression (PLSR) model was developed using calibration and validation sets for the spectral range of 950-1650 nm and a separate set was used for obtaining prediction accuracy.

A. OPTICAL DESIGN AND SIMULATION ANALYSIS
Using the Zemax non-sequential mode, an ELICO NIRS prototype and an NIRscan Nano EVM simulation model was designed to illuminate the sample at a specified standoff distance. A one-watt per lamp source was chosen for a standard model of NIRscan Nano EVM and two-watt per lamp source for the prototype of ELICO NIRS model. Sources were positioned at an optimum location and angle to achieve the necessary overlap region to illuminate the target sample position. We match the overlap of the active region and the optics vision cone by selecting the appropriate optics with "f/number." A detector was placed at the sample location to determine radiometric output. This metric calculates the amount of illuminating power that reaches the active region. The physical properties of the lamps and the optics of the illumination module were redesigned and optimized. The design must improve to obtain the desired level of illumination power. The design was deemed satisfactory when the preferred metric reached maximum power at the target sample location. In Zemax non-sequential mode, accurate design analysis requires selecting the correct source, power distribution, and detector type. The source model determines the cone angle and energy distribution of the light. Since the "detector rectangle" version is helpful for our analysis, there are several options for the detector type. This method allows the user to determine the number of pixels and sizes on the sensor. For an accurate study, the NIRscan Nano EVM's detector size is 2.5 mm, and the ELICO NIRS prototype model's detector size is 4 mm. The system design was set up in Zemax for analysis and run ray-trace simulation. After completing the ray trace, the irradiance at the detector plane was measured. The ray trace uses the detector viewer, allowing for different irradiance views. The resultant illumination module ray geometry of both models is shown in Fig. 3. The results from the Zemax simulation of various standoff distances against the sapphire window and the corresponding overlap area and illumination power are shown in Fig. 4.
As shown in Fig. 4, NIRscan Nano EVM provided a 2.5 mm SAIA and 66 percent light power to illuminate the sample at '0 mm' position, allowing for accurate analysis. Because of the shallow depth of illumination, only 45 percent of the light energy was detected at a distance of 1 mm from the sapphire window, which was insufficient for precise measurements. Because of this geometry, the sample can be used for measurements up to 0.75 mm. The ELICO NIRS Prototype model increased SAIA by 4 mm and DOIR by 2 mm. Illumination power was 72 percent at '2.5 mm' and 62 percent at '4.5 mm,' and this range was used to keep the sample for reliable analysis.    accuracy. Fig. 3 shows the ray geometry of an illumination module, illustrating the geometric variations of the proposed model and the NIRscan Nano EVM.

C. MULTI-POINT SCAN WITH ELICO NIRS PROTOTYPE
The ELICO NIRS prototype consists primarily of a newly developed illumination module, the NIRscan Nano EVM's polychromator, and a microcontroller that allows for multipoint scanning of the bulk sample at various points while rotating the sample holder. The proposed geometry of the illumination module allows for 3.5 mm of space for the rotation mechanism, sample holder, and sample. As a result, the measuring device illuminates the sample sufficiently and takes several measurements as the sample holder rotates, ensuring that the measurement is as homogeneous as possible.

D. PERFORMANCE OF ELICO NIRS PROTOTYPE
The prototype's performance is validated in terms of throughput, baseline, SNR, repeatability, linearity, and accuracy.

1) THROUGHPUT (INTENSITY)
Throughput was measured by using a standard diffuse reflectance. The scan was performed to obtain NIR spectrum scan of a known diffuse reflectance standard on NIRscan Nano EVM and ELICO NIRS Prototype. Fig.  5(a) shows the output intensity curves for both models. The abscissa is the wavelength, and the ordinate is the relative power strength in analog units. The output intensity of NIRscan Nano EVM is red, while the output intensity of ELICO NIRS prototype model is blue. In comparison to the output intensity of both models, the proposed ELICO NIRS prototype geometry increases the resultant intensity by a factor of 2-3, as shown in Fig. 5(b).

2) BASELINE
A baseline test scan is performed at various gains across the entire wavelength spectrum and compared for both instruments. Finally, the corrected signal is evaluated for peak-to-peak fluctuations. The NIRscan Nano EVM achieved <± 0.005Abs in all gain settings, whereas ELICO NIRS prototype achieved <± 0.003Abs at gain-1 and <± 0.001Abs at gain-16 & 64, and the results are shown in Fig.  6 (a, b, & c). Finally, ten consecutive baseline scans were performed, and the data was captured using ELICO NIRS prototype model. All the baseline scans within <± 0.003Abs as shown in Fig. 6(d).

3) SNR
SNR calculations were performed on ELICO prototype and NIRscan Nano EVM, with 100 scans captured on both models at 15 ms, 30 ms, and 60 ms integration times, and SNR was determined using the formula below, Table 2 shows that the proposed ELICO NIRS prototype geometry increased throughput, baseline, and SNR compared to NIRscan Nano EVM. In addition, further study showed that the Limit-of-Detection (LOD) and instrument sensitivity had greatly improved.

4) REPEATABILITY
Repeatability experiment was performed for inter-day and intra-day and this was evaluated to determine how repeatable instrument results are under a set of similar conditions. Repeatability was determined by analyzing pure powder and grain samples' spectral data using the multi-point scan method. The result % amount found was between 99% and 101% with %RSD <1% by applying MSC. As shown in Table-3, results indicate that this prototype is precise for the determination of the quality of the food samples using multi-point scan method.

5) LINEARITY & ACCURACY
Poultry feed samples with moisture in the range of 10 to 20% were scanned on an ELICO NIRS prototype using a multi-point scan technique. As shown from the original spectrum in Fig. 7, there is a serious drift, although the spectra are similar overall. The raw spectra were subjected to Savitzkey-Golay smoothing, SNV or MSC, and first derivative to establish a robust quantitative model. These techniques increase contrast and aid in the resolution of overlapping moisture content peaks. Fig. 8 depicts the resulting linear variation at 1450nm for each moisture level. To compute the moisture content of the feed sample, we applied raw spectral data, well-pre-processing techniques, and partial least square regression (PLSR) to develop better calibration and validation statistical prediction equations. Correlation coefficient (R 2 ), cross-validation root mean square error (RMSECV), prediction root mean square error (RMSEP) were used as evaluation criteria. The smaller the RMSE value and the closer R 2 value to 1, the greater the correlation between the real and predicted values. Table 4 shows the results of different pre-process techniques after PLSR modeling. The results show that the model that uses the first derivative is better than other pre-process techniques. Fig. 9 shows the resulting moisture content of each sample in the calibration set plotted against the wet chemistry values. The moisture content prediction model produced had an excellent R 2 of 0.9915, RMSECV of 0.247, and RMSEP of 0.324 percent.  The evaluation parameters are reasonable, so this study used modeling result as the final result. The prediction set samples were predicted using the above model, and the predicted results were compared with the standard wet chemical values, as shown in Table-5. It can be seen from the results that the maximum percent of deviation of the system is less than 1%, and the minimum is 0.3%. Thus, in this study, ELICO NIRS prototype shows that it has high prediction ability for multi-point scan analysis and can be applied for rapid detection of standoff distant heterogeneous samples and is suitable for use in labs of food industries where users can obtain results quickly and accurately.

V. CONCLUSION
This paper presents ELICO NIRS prototype illumination module using the NIRscan Nano EVM to perform multipoint scan analysis of standoff distant samples. The proposed module's geometry improved optical characteristics in terms of SD, SAIA, and DOIR. The standoff distance (SD) achieved is about 3.5 mm, allowing standard sample holders to retain the sample for a single-point scan and a sample rotation mechanism for multi-point scan analysis. The prototype model achieves an SAIA with a maximum intensity of 4.0 mm at a standoff distance of 3.5 mm from the sapphire window by improving optical geometry. It improves detectivity by a factor of 2 -3 by collecting maximum diffused reflection light from the sample by proper collection optics. The DOIR increased to 2 mm, allowing the sample's position without compromising the accuracy of the analysis. The evaluation results show a noticeable improvement in instrument baseline of around <±0.003 Abs at gain 1 and <±0.001 Abs at gain 16. The study's findings revealed a low-cost portable instrument with improved signal to noise ratio (SNR) and limit of detection (LOD), as well as the ability to conduct multi-point analysis on bulk, inhomogeneous, and heterogeneous samples. PLSR analysis on a data set of feed samples with moisture ranges of 10-20% percent yielded a prediction model with acceptable precision, with an R2 of 0.991 and a prediction error of 0.32 percent. ELICO NIRS prototype prediction values of a feed sample highly correlated with wet chemistry method's values, with a maximum deviation of <1%. Therefore, it can be concluded that the ELICO NIRS is sensitive and is ready to read distinctive variations levels in the feed samples in our lab setup. Since the ELICO prototype is cost-effective and generates results rapidly, feed manufacturers can maintain tight control and monitor what is going into the finished feed, thereby reducing production losses during the manufacturing process. According to this section of the fundamental study, the ELICO NIRS prototype combined with chemometrics has vast potential as a stand-alone or in-line monitoring tool for food applications.