Correlation between oil content and DN values

Correlation between oil content and DN values

Rashid Shariff, Nor Aizam Adnan, Radzali Mispan, Shattri Mansor, Rohaya Halim, and Roop Goyal
Department of Biological & Agricultural Engineering, University Putra Malaysia, Malaysia
rashid@eng.upm.edu.my

Abstract
This research investigates the correlation between oil content in the oil palm fruit against the color of the oil palm fruit. Images from samples of oil palm fruits were acquired using a camera with proper lighting and angle. Image processing and analysis of the samples was carried out. Physical measurements of the amount of oil content in the fruit were carried out by chemical analysis. This study revealed that there is a high positive correlation between the color of the oil palm fruit and the amount of oil in the fruit. This study will be of help in increasing the efficiency of quality harvesting and grading of oil palm fresh fruit bunches (ffb).

Introduction
Currently, oil palm fruitlet quality is graded manually. This method has some disadvantages. Firstly, it is an extremely tedious and time consuming process prone to errors or inconsistencies. It is inaccurate and has a strong bias towards the mill. So a new technique is needed in grading oil palm. This research project investigates the correlation between recognisable parameters of oil palm fruit such as color against the content of the oil palm fruit. The findings of this study will lay the foundation for better oil palm grading system.

Table - 1: Fresh Fruit Bunch Standard

Category	Description
Ripe	10%-50% of fruits detached from bunch
Over ripe	50% - 90% of fruits detached from bunch
Under ripe	1 - 9 fruits detached from bunch

Source: MPOB, 2000

Fig 1: Ripe

Fig 2: Over ripe

Fig 3: Under ripe

This study is confined to only a single species of oil palm fruit. The Tenera species of oil palm fruit was used for this purpose because it has a thick mesocarp. Samples of this study are based on three kinds of ripeness of fruits which are known ripe, under ripe and over ripe. This study was conducted in a room that had proper lighting technique and with known distance between the camera and the fruit.

The images covered the side view of the fruit and does not include the bottom of the fruits. This study did not involve the analysis of texture and shape. The focus of this study is the fruitlets and does not cover oil palm bunches

Materials and Methods
Fruitlets sampling is carried out because oil palm fruits have different ripeness stage. The oil palm fruits can be ripe, unripe, overripe, black bunch or empty bunch. For this study we used 3 categories of oil palm fruits, which are known as ripe, unripe and over ripe because this ensures a good cross- selection across the board (Table 1). A Nikon macro-lens camera was used to acquire image of these fruitlets. A total of 30 samples, made up 10 samples each of ripe, over ripe and under ripe fruitlets ( Fig. 1, 2 and 3 ). Then images of these fruitlets were captured in a room with a proper control environment such as lighting, distance and height from camera to object. The setting of this image acquisition system is shown in Fig. 4.

This study involved hardware system in order to capture an image, process the data, display the image and to do statistical analysis. The main hardware system in this study is the camera to grab the image of the sample oil palm fruits. The camera used in this study is the Nikon CCD RGB Single Chip and the computer system to do image processing and statistical data analysis is the Intel Pentium III processor, 64 MB RAM. The pictures were scanned to convert them into digital form using a scanner with 400 dpi optical resolution.

PCI 7.0 application software was used for the digital image processing and analysis. Operating system used was Window 98 system. PCI 7.0 software is capable of image displaying, enhancement, manipulation and restoration.

Information that is extracted in this study is the color value based on the RGB color model. The image processing is done to see the relationship of the color with oil content of oil palm fruit. By using PCI software we can see each band (Red, Green and Blue’s color in the 3 types of fruits with different ripeness.) Then for each of these colors we get the Digital Number (DN) for each band. The information is used to plot the histogram of each band for each type of fruit (Fig. 5 (a), (b), (c)).

In order to investigate the correlation between color properties with oil content, chemical processing was carried out to determine the oil content for every sample. In doing this analysis, lab processing is done immediately after the image of the oil palm fruits is grabbed using camera. This is because the oil content might change if we process the samples later than one week. Formula to measure the oil content is shown below:-

% Oil = Wt of Oil X 100% Wt sample (1)

Where:

Wt oil = weight of oil
Wt sample = weight of sample

Fig 4: Image Acquisition Setting

The oil content that has been extracted using this chemical processing is determined as a percentage (%) of the weight of the sample. This analysis is repeated until the entire sample is done. From these results, we can correlate the oil content values by comparing against the color values of the three types of fruits that were used.

**Table 2:** *Chemical analysis results*
Sample	Ripe (%)	Over Ripe (%)	Under Ripe (%)
1	13.88	10.16	4.68
2	6.77	13.46	5.85
3	17.32	16.55	4.92
4	17.82	15.91	7.83
5	19.63	14.89	6.76
6	9.18	14.84	7.67
7	14.47	14.17	9.34
8	20.45	13.18	6.68
9	10.06	16.38	8.90
10	26.84	20.40	8.96
Mean	15.64	14.99	7.16

Fig 5(a): RGB for Ripe Fruit Sample

Image Processing
Image processing in this study used graphic mask operation and multi plane histogram. Graphic mask operation objective is to get the value of DN within the fruit excluding the external area. This analysis gives an accurate DN value of the samples that was used because it eliminates shadows and objects that are not related to this study

Fig 5(b): RGB for Over Ripe Fruit Sample

Results

Chemical Analysis
A chemical analysis was done to determine the amount of oil content in the oil palm fruit. The results from the lab chemical processing of these three different kinds of fruit is shown in Table 2

**Table 3:** *DN Values for Ripe, Over Ripe and Under Ripe Fruits*
CATEGORY	RIPE			OVER RIPE			UNDER RIPE
Sample/Band	R	G	B	R	G	B	R	G	B
1	135	18	6	183	32	14	156	81	16
2	102	21	6	179	37	16	113	49	5
3	140	42	18	184	52	29	147	60	7
4	155	42	14	185	50	26	128	53	8
5	157	43	17	193	55	20	120	51	12
6	124	27	7	195	44	19	152	68	7
7	144	30	8	185	60	17	188	98	20
8	159	28	10	174	33	18	150	68	18
9	139	33	8	166	44	24	160	64	4
10	172	60	23	203	69	29	152	60	8
Mean	142.7	34.4	11.7	184.7	47.60	21.20	146.6	65.2	10.5

Fig 5(c): RGB for Under Ripe Fruit Sample

RGB Colour Analysis
Ripe oil palm fruit is reddish in color. In this study, a camera captures the image of the fruit by using 8-bit storage which contains 3 planes known as Red, Green and Blue planes. Combination of these three planes makes it 24-bit. Each plane gives different DN value with reference to the fruit. Table 3 shows DN value results for ripe, over ripe and under ripe fruits sample.

For the Ripe fruit category, the red band shows the highest value compared to the Green and Blue bands. This is because the ripeness of oil palm fruit is indicated by the reddish color of ripe fruit. For the green band, the mean DN value is in the range of 18 < = Green_ripe > = 60. For the blue band, the range is < = 6 Blue_ripe 23. < = Red band of over ripe fruit shows the highest DN value. The range of mean DN values is 166 < = Red_{over ripe}203. < = The oil content for this category is between 10.16% and 20.4%. For green band, the range of mean DN values is between 32 < = Green _{over ripe} < = 69. For the Blue band, the range of DN values is between 14 < = Blue_{over ripe} < = 29. Although the RGB values for this fruit samples shows higher value compared to the ripe fruit, the mean oil content is approximately the same compared to the ripe fruit sample. Mean DN values for the under ripe fruit category is in the range of 113 < = Red _{under ripe} < = 188 for the red band, for the green is 49 < = Green _{under ripe} < = 98 and for the blue band the range is 4 < Blue _{under ripe} < = 20. The mean oil content of the under ripe fruit is 7.16%.

Fig 6(a): Linear Regression of Red Band for Ripe Fruit Sample

Regression Analysis & Accuracy Assessment

A linear regression was carried out using the oil content and the DN values. Figure 6(a) shows that, for the Red band, there is good correlation between the oil content and the DN values.

The correlation is represented by value R² = 0.8549. For the green and blue band, values for R² are 0.6026 and 0.704 respectively. These values show good correlation between this two parameters as the value of R² approaches 1.0 (Fig. 6 (a), (b), (c)).

Fig 6(b): Linear Regression of Green Band for Ripe Fruit Sample

For the over ripe fruits, the regression line is shown by the Fig. 7 (a), (b), (c). From this line, we can conclude that the blue band gives the highest correlation which can be seem by the figure R² = 0.7832. The red band is represented by R² = 0.1793 and the Green band R² = 0.6293. From this figure, we can conclude that for the over ripe fruit, Blue band is significant to estimate oil content.

Under Ripe fruit category does not show any significant relationships between DN values and oil content in estimating the oil content. This can be seen from the regression line of this fruit. The R² value for each red, green and blue band is 0.2347, 0.1253 and 0.0925 respectively. From a statistical view, we can conclude that color analysis for this fruit category cannot be used in estimating oil content. (Fig. 8 (a), (b), (c)).

Fig 6(c): Linear Regression of BlueBand for Ripe Fruit Sample

To test this correlation result, accuracy assessment was carried out. The accuracy assessment is shown in Table 4. For the ripe fruit, the mean accuracy for the red band is 87.41% while for the green and blue bands the mean accuracy is 84.04% and 82.83% respectively. For the over ripe fruit, the mean accuracy for the red band is 88.63%, the green band is 92.07% and the blue band is 93.73%. For under ripe fruit, the results are between 82.13% up to 84.13%.

The correlation analysis results can be classified qualitatively. The ripe fruit category can be represented by a combination of “high” correlation values for the red band, while green and blue bands have “medium” correlation values (Table 5). For the over ripe fruits category the red band has “low” correlation values, while green and blue bands shows “medium” correlation values. For under ripe fruit category, red, green and blue bands shows “low” correlation values. This qualitative classification can lead to a unique differentiation of the different ripeness categories.

Fig 7(a): Linear Regression of Red Band for Over ripe Fruit Sample

Fig 7(b): Linear Regression of Green Band for Over ripe Fruit Sample

Fig 7(c): Linear Regression of Blue Band for Over ripe Fruit Sample

Conclusions
This study was conducted to study the correlation between color properties of the oil palm fruit and the amount of oil content within the fruit. From the analysis, we found that this study achieved its main objective of testing the use of digital image processing to estimate oil content of the oil palm fruit.

Results show that for the ripe fruit samples, red band shows the best correlation with oil content compared to blue and green band. For over ripe fruit, blue band showed a good correlation with oil content. This allows us to differentiate between the ripe and over ripe fruits. For the under ripe fruit there is no significant relationship with oil content. This is true because under ripe fruit are not matured as yet to give good oil content. Overall, we can conclude that, for ripe fruit category red band is very suitable to be used as a parameter to estimates oil content. For over ripe fruit the blue band can be used.

We found that the mean DN values are not directly able to differentiate the amount of oil content between the ripe and over ripe oil palm fruitlets. However, correlation analysis between the DN values for the three RGB bands against the % of oil content was clearly able to differentiate between the ripe, over ripe and under ripe fruit categories. This is both an important and significant finding as a distinction between the three categories of ripeness of the fruits can be determined. Although the % of oil content in the ripe and over ripe category does not show a substantive difference, the quality of oil in these two categories of fruits is substantially different. Thus, the ability to differentiate between these two categories of ripeness using image analysis technique as demonstrated in this research will be of help in quality grading of fruits in the oil palm industry.

Fig 8(a): Linear Regression of Red Band for Under ripe Fruit Sample

Fig 8(b): Linear Regression of Green Band for Under ripe Fruit Sample

Fig 8(c): Linear Regression of Blue Band for Under ripe Fruit Sample

References

Awcock G.J , Thomas R, 1995, Applied image processing, London: MacMillan Press Ltd.
Baxes Gregory A., 1994, Digital image processing: principles and applications, John. Wiley & Sons.
Chen Rud-Ren, 2000, Future trends of machine vision technology for agricultural applications, USA:Betsville Agricultural Research Center .
Daley, W., et. al., 1993, Poultry grading/inspection using color imaging, SPIE Vol. 1907:124-132,
Delwiche, M.J., et. al., 1987, Color and optical properties of clingstone peaches related to maturity, Trans. ASAE, Vol. 30(6), Nov-Dec 1987.
Garcia-Alegre Maria C. et. al, 2000, Color index analysis for automatic detection of eggshell defects, Madrid, Spain : Spanish Council for Scientific Research..
Marimoto T., Takeuchi T.,Hashimoto Y, 2000, Pattern recognition of fruit shape based on the concept of chaos and neural network, Journal of Computer and Electronics in Agriculture. 26 (2000): 171-186.
MPOB,2000. Manual grading of fresh fruit bunch
Rehkugler, G.E., and J.A. Throop, 1989, Image processing algorithm for apple defect detection. Trans. ASAE . 32:267-272.
Throop, J.A., D.J. Aneshansley, and B.L. Upchurch. 1993. Near-IR and color imaging for bruise detection on Golden Delicious apples, Proc. SPIE 1836:33-44.
Wan Ishak Wan Ismail, Mohd Zohadie Bardaie & Abdul Malik Abdul Hamid, 2000, Optical properties for mechanical harvesting of oil palm ffb. Jounal of Oil Palm Research. 12: 38-45
Li Qingzhong, Wang Maohua, 2000, A machine vision system for apple surface defect detection, Information and Intelligent Institute, Zhejiang University. China.
Tao, Y., et. al., 1995, Machine vision for color inspection of potatoes and apples, Trans. Of ASAE, 38(5) : 1555-1561
Zuech, Nello, 1988, Applying machine vision, John Wiley & Sons.