ON-THE-GO SENSING FOR SUGAR CONTENT DETERMINATION OF SUGARBEETS IN THE FIELD

Suranjan Panigrahi, Asst. Professor of Agricultural Engineering
Vern Hofman, Extension Ag Engineer, Ag & Biosystems Engineering
Dan Gu, Graduate Research Assistant

Introduction:

Sugarbeet is a crop of economic importance to the region. Determination of sugar content of the crop in the field is an important process for optimized planning and production. Determining the sugar content along with the tons of beets which is already being measured will allow producers to determine the productivity in pounds of sugar.

The present technique used for determination of sugar content involves a laboratory procedure. Any means of determining the sugar content in the field on-the-go could be beneficial in improved management of the sugarbeet crop in real time. Thus, it would contribute to the profitability of the sugar industry.

In recent years, the emergence of advanced sensing and sensor technologies such as fiber optics and near infrared technologies, have shown potential to develop efficient, reliable, compact and cost-effective sensors/sensing processes for agricultural and food processing applications. Moreover, the concept of precision farming with the incorporation of Global Positioning Systems (GPS) promises efficiency, sustainability and profitability for modern farming including the sugarbeet crop. The availability of reliable, compact and on-the-go sensors are essential to harness the maximum benefit from precision farming. For example, the integration of an on-the-go sugar content sensor with GPS and a yield monitoring system could provide benefits for cost-effective management and production of sugarbeets.

Literature Review:

Near-infrared (NIR) technology has been used as a quick, reliable, and non-destructive means for determining different constituents (protein, oil, starch, moisture, carbohydrate, sucrose, etc.) of a variety of food and agricultural products. For all these measurements, generally a laboratory scale NIR spectroscopy in reflectance or transmission mode has been used. (Osborne and Fearn, 1986).

Recently, there has been a need for an on-line or real-time measurement for food processing applications. NIR instruments (spectroscope) are being used (Baldwin, 1994). NIR techniques have also been used for quality control of sugar factories (Clark et al, 1994). Though these investigations prove the capability of the technology for the on-the-go determination of sugar content in the field, no research work has been found to apply the proposed sensor on a beet harvester.

Calibration is a critical component of NIR-based analysis system (Baldwin, 1994). To date, statistical techniques have been used for developing the calibration model for NIR analysis. However, statistical techniques have some limitations which are easily handled by another new form of information processing technique called "Neural Networks". Neural networks process the information in a parallel and distributed form, analogous to that of a human being. Thus, neural networks are more fault tolerant and robust with the capability to learn from its own error. Neural networks have been used for many different agricultural applications, ranging from disease prediction and weather forecasting to farm management and quality control operations (Panigrahi, 1995). Thus, neural network techniques will be suitable for developing an on-line calibration model for NIR based analysis system.

Objective:

The objective of this research is to develop an on-the-go sensor for the determination of sugar content of sugarbeets in the field during the harvesting operation.

Procedure:

A PC-based fiber optic spectral meter (800-1700nm) was used to acquire the NIR transmission (absolute) signal. Initially, experiments were conducted to optimize different optical configurations for acquiring the signal. A customized sensor head consisting of achromatic lenses was used. A specialized tungsten halogen lamp (100 Watt) was used as the light source.

A sample holder was specially built using high transmission transparent PVC. This sample holder was used to hold 5mm thick 2" x 4" rectangular cross section of sugarbeet. Nineteen sugarbeet roots were collected at random. And for each root, the skin was peeled off. The required rectangular section of the sugarbeet was taken out from the upper one-third of the beet root. Their absolute NIR transmission signals were acquired and stored in the computer. Then the same sugarbeet cross sections were analyzed for their actual sugar content using conventional wet chemistry method.

The acquired NIR signals were further processed using different statistical techniques such as averaging and second derivatives. The processed signals contained 122 data points or inputs. These inputs were fed to different statistical prediction models for predicting their sugar contents. All the inputs were normalized with mean of zero and standard deviation of one.

Results:

Out of different prediction models, principal component and partial least square methods provided the best prediction accuracy. For principal component method, 18 principal components were selected to be optimal. Using these 18 components, the mean square error of prediction of sugar content was 1.08 x 10-7. The average absolute error was 2.47 x 10-4. And the minimum and maximum absolute error in predicting sugar content was very low. Figure 1 shows the comparisons of actual and predicted sugar content of these sugarbeets. Table 1(a) shows their corresponding represented values. This shows that the technology used has potential in predicting sugar content with very high accuracy.

Partial least square method (Figure 2) used 14 factors to predict sugar content of the sugarbeet with a mean square error of 1.45 x 10-10. The average absolute error was 1.05 x 10-5. The minimum and maximum absolute error in predicting sugar content was also very low. Figure 2 shows the comparisons of actual and predicted sugar content of these sugarbeets. Table 1(b) shows their corresponding represented values.

Though partial least square method shows relatively lower error, both methods have very low errors for all practical purposes. Both methods show high accuracy in predicting sugar content of the sugarbeet.

Conclusion:

From this study, it was concluded that using NIR spectralscopy in 800-1700nm range along with suitable statistical prediction model can predict sugar content of the sugarbeet with very high accuracy. This study shows that this non-destructive method can be used for prediction of the sugar content of the sugarbeet in a fast and accurate manor.

Future Work:

A follow-up study will be conducted for further validating the model on a large number of sugarbeet samples. Additional study will also be conducted to evaluate the NIR reflectance technique for prediction of sugar content. Though this study shows the feasibility of NIR transmission technique, it would not be suitable for determination of sugar content in the field. However, NIR transmission technique can be used for determination of sugar content in post harvest condition. It is anticipated, based on the study, that NIR reflectance technique can also show very positive results.

References:

Osborne, B. and T. Fearn. 1986. Near Infrared Spectroscopy in Food Analysis. John Wiley and Sons, New York, NY.

Baldwin, E. 1994. Calibrating NIR instruments for on-line food processing measurements. Proceedings of the 1994 Food Processing Automation Conference. Orlando, FL.

Clarke, M., L. Edye, C. Scott, X. Miranda, and C. McDonald-Lewis. 1992. NIR Analysis in Sugar Factories. Proceedings of Sugar Process Res. Conference. New Orleans.

Panigrahi, S. 1995. Neuro-fuzzy systems: Their potentials and applications in biology and agriculture. Accepted for publication in AI Applications.

Figure 1 Principal Component Method with 18 Components

Figure 2 Partial Least Square Method with 14 Factors

 

Table 1 Actual and Predicted Sugar Contents
using Different Statistical Prediction Models
Sample # Actual Predicted Sample # Actual Predicted
1 0.1933 0.193064 1 0.1933 0.193283
2 0.2101 0.210056 2 0.2101 0.210091
3 0.2044 0.204345 3 0.2044 0.204407
4 0.1946 0.195088 4 0.1946 0.194607
5 0.1973 0.196725 5 0.1973 0.197285
6 0.1925 0.19241 6 0.1925 0.192505
7 0.1779 0.17854 7 0.1779 0.17791
8 0.2084 0.208303 8 0.2084 0.208394
9 0.1874 0.187083 9 0.1874 0.1874
10 0.2232 0.223442 10 0.2232 0.223222
11 0.2073 0.207202 11 0.2073 0.207301
12 0.1673 0.167314 12 0.1673 0.167319
13 0.1719 0.171661 13 0.1719 0.171886
14 0.1983 0.198501 14 0.1983 0.198292
15 0.1941 0.194032 15 0.1941 0.194109
16 0.1837 0.184374 16 0.1837 0.183713
17 0.2074 0.207395 17 0.2074 0.207384
18 0.1882 0.187672 18 0.1882 0.188185
19 0.2217 0.221792 19 0.2217 0.221708

                                                             (a) Principal Component Method                                  (b) Partial Least Square Method
                                                                              with 18 Component                                                with 14 Factors


1998 Sugarbeet Research and Extension Reports. Volume 29.