Analysis of Physical Properties Of Oil Palm Fresh Fruit Bunches Using ImageJ

Transcription

1 ICST 2016 Analysis of Physical Properties Of Oil Palm Fresh Fruit Bunches Using ImageJ Minarni Shiddiq 1*, Roni Salambue 2, Rasmiana Poja 1 and Arian Trianov Solistio 1 1 Department of Physics, Universitas Riau, Indonesia 2 Department of Mathematics, Universitas Riau, Indonesia minarni@unri.ac.id, roni.salambue@unri.ac.id, rasmianasiregar@gmail.com, and ariantrianov@yahoo.com *Corresponding Author Received: 8 October 2016, Accepted: 4 November 2016 Published online: 14 February 2017 Abstract: Computer vision methods has been developed to assess the quality of oil-palm fresh fruit bunches (FFBs). Most researches evaluate the FFB qualities using soft computing techniques which need computer programming skill. A simple method using ImageJ software was proposed to analyze the physical properties FFB with different ripeness stages. The FFB images were recorded using a USB CMOS color camera. The physical properties include red green blue (RGB) intensity, number of outer fruits, length and equatorial diameter, and estimated density of a FFB were obtained using ImageJ software and compared to the results measured manually for front and backside of FFB images. The samples were oil palm FFBs from Tenera variety consisted of two types named Tenera A and Tenera B. Each type has 4 ripeness stages with 3 FFBs for each stage. The results show the average RGB intensities of Tenera A are slightly higher than Tenera B for all ripeness stages due to bigger fruit size. Percentage differences of physical properties measured by both methods range from 1% -15 % except for over ripe FFBs due to irregularity in fruit number and density. ImageJ software is shown to have the capability analyzing the physical properties of the FFBs. Keywords: Computer vision; oil palm fresh fruit bunch; ripeness; physical properties; ImageJ. 1. Introduction Sorting and grading oil palm fresh fruit bunches (FFBs) are two important procedures performed before the FFBs sent into crude palm oil (CPO) milling places. Most sorting and grading oil palm FFB methods are based on the FFB ripeness stages and qualities. FFB ripeness is usually classified by its color or by number of detached fruits found on ground. Qualities of a FFB are measured by biochemical content such as oil contents and free fatty acid (FFA) contents, and physical properties such as shapes, textures, lengths, and density. Until now, sorting and grading oil palm FFBs are completed manually using human vision by experience harvesters and graders. Sorting and grading using traditional ways are subjective, time consuming, laborious [1]. Sorting and grading using reliable methods are needed because of many reasons such as reproducible, more accurate, objective, can be non destructive and fast [2]. Razali et al. [3] have reviewed many methods that have been developed for sorting and grading oil palm fresh fruit bunches (FFBs) based on ripeness and qualities. Researches in this field are driven to develop a nondestructive, simple, fast, reliable, cost-effective system for harvesting, sorting, and grading that can be used by individuals or oil palm companies. Computer vision is one of sorting and grading methods that have been intensively proposed for oil palm FFBs. Computer vision is used to imitate the functions of human vision using computer hardware and software [4] which sometimes is called machine vision because they are used in industries to sort or grade products. This method is also known as visible imaging techniques because images are acquired in the process. For oil palm FFBs, the methods are used based of oil palm fruit

2 color to classify ripeness stages because the FFB color changes during ripening, the ripeness are classified using color digital number or red green blue (RGB) values [5,6]. There are at least 4 processes performed in computer vision i.e. image acquisition, image processing, analysis, decision making. The main component of the computer vision is a digital camera with lens used to acquire images from objects. The camera comes in two types, a commercial camera and a scientific camera. Commercial cameras (fast camera) are found in cellphones and photographic cameras while scientific cameras are found in scientific equipments. The cameras have two types of photosensor, CCD (charged coupled devices) and CMOS (complementary metal oxide semiconductor) which have advantages and disadvantages. Image processing is very important part in computer vision. Some researches focus on developing image processing program which require computer programming skills [7]. Other researches focus on the results or other things. The latter likely uses commercial image processing software, however the software can be expensive but sometimes is open-source. ImageJ software is a non- commercial software which is widely used in scientific researches. ImageJ is a Java-based open source software which was developed for first time to analyze images at National Institute of Health (NIH), USA. At first, it was written for Macintosh based images, latter in 1997, the coauthor of the software expanded it in Java programming language which can be available and used by everyone. ImageJ can work at any operating system. The uniqueness of this software is that Wayne Rasband, the coauthor, only write the program core, many groups have written short -add on program (called plugins) intended to increase its functionality for many image processing problems. There are more and more plugins available to download every day and can be loaded and used in the core program [8]. ImageJ software is widely used in Biology and Medicine for images obtained using digital microscopes [9, 10]. ImageJ has been used in Science and Engineering for material analysis, recently are used for fruit qualifications [11]. Qualities of fruits can be described by their physical properties. Physical properties that can be used to classify and sort fruits are color, size, shape, mass, firmness, or damages on the fruits. ImageJ can be used to classify or to grade the fruits based on their physical properties [11]. Qualities of fruits can be measured using their physical properties that can be detected by human senses such as color, texture, and shape by eyes, taste using tongue, smell or senescence by nose, and firmness by touching [2]. The human senses have been replicated using electronic sensors such as digital camera and electronic nose. The purposes of this replication are to create a non destructive, reliable, low cost system to classify, sort, grade fruits and vegetables. Oil palm fresh fruit bunches (FFBs) have physical properties such as color (optical), shape, size, mass that can be used to classify them in order to get good quality FFBs which are marked by optimum oil content and low FFA content [12]. ImageJ can be used to grade the FFBs based on their physical properties. In this present study, the physical properties of FFBs of Tenera variety, color, and number of outer fruits, length and equatorial diameter, estimated density are obtained using ImageJ software v. 1.47c from FFB images acquired by a CMOS camera. The results are compared to results measured manually. The parameters compared are number of outer fruits, length and equatorial diameter, estimated density. The remainder of the paper is organized as follows. In Section 2, classification methods of ripeness stages of oil palm FFB and capabilities of ImageJ program for fruits classifications will be described. Section 3 explain the material and methodology used, Section 4 will describe the results and followed by discussion. Finally, we describe conclusion and future work in Section Literature Review FFB ripeness can be classified by many ways including by its color or by number of detached fruits fall to the ground for the FFB. Color FFB depends on variety. Based on color, Oil palm FFBs are categorized as Nigrescence and Virescence. For Nigrescence type, FFB color changes from dark puple to dark red and orange while for Vigrescence changes from green to dark orange during ripening [1]. In general, there are 7 ripeness stages which sometimes called fraction of FFB based on the loose fruits. They are F00 (unripe1), F0 (unripe2), F1 (under ripe), F2 (ripe 1), F3 (ripe 2), F4 (overripe1), F5 (overripe2). In palm oil millings, the ripeness is classified by 3 categories, under ripe, Applied Science and Technology, Vol.1 No

3 ripe, and overripe [13]. Physical properties of FFB are different for different ripeness stages. Back side and front FFB image should be analyzed because their fruit density is different. Front side FFB is the side facing sunlight which has homogenous fruit density even though the fruits size slightly vary at different section of a FFB. Back side FFB is the side that attached to fronds. It has less fruits, and the size is extremely different at three section of FFB (basal, equatorial, apical) [3]. At basal section, the fruits density is high with very small fruit size and contain the least oil. Researches for ripeness and quality classification of FFBs have been done extensively in last 10 years. Qualified FFBs are FFBs which give maximum oil contents [3]. There are numerous methods have been developed. A compact grading system has been developed using a computer based photogrammetric method. In this system, white diffused light was used in a room light tight box. FFB Ripeness was classified using color Digital Number (DN) [14]. An automatic FFB grading system was also developed using computer vision method. This system could estimate the ripeness fraction, oil content, and fatty acid content of FFBs. Two statistical methods have been used, forward stepwise multiple linear regression analysis and multilayer-perceptron artificial neural network [15]. These methods are based on computer vision and white light excitations. Images of FBs are processed to extract gray values or RGB values. Models in classification of FFB ripeness have also been developed to create a decision making software. 3. Material & Methodology 3.1. Sample Preparation This study was aimed to use computer vision method and ImageJ software to analyze the relations between the physical properties of oil palm FFBs and ripeness stages. Samples of this study were oil palm fresh fruit bunches from Tenera variety which consist of two types which grown in most Indonesian oils palm plantations, they are called Tenera A and Tenera B. Both have distinctions in bunch size, fruit size, and fruit density or number of fruits in a bunch. The samples consisted of FFBs with 4 ripeness levels called fraction, F2and F3 (ripe) and F4 and F5 (overripe) for Tenera A and F0, F1, F3, F4 for Tenera B. Number of fractions found during harvesting depends on situations, only 4-5 fractions are found harvested which can be sold at oil palm mill. Fraction below F1 are rarely harvested. The classifications of the ripeness were done when they were harvested by an experience harvester. Each level had 3 FFB duplicates. The FFBs were taken from oil palm plantations which grown less than 10 years. After harvested, the samples are cleaned from dirt and physical properties i.e. mass, length (distance from top of apical section to end of basal section) [3], equatorial diameter, number of outer fruits. The diameter and length were measured using rolling ruler wrapped to the FBB and the result then were divided by 2. The number of fruits were calculated for all outer fruits that can be counted but not included very small fruit at back basal section. All the measurements were done three times with different person but the same methods, and then the results were the averages of the three measurements. The discrepancy in the measurements could be due to variations of size and mass for each sample, for each type Method The computer vision system used for FFBs consisted of a CMOS camera and one set of computer. The optical setup was stated on a stainless steel optical post for easy height and position adjustment. The USB-CMOS camera is positioned at 2.07 m from an aluminum platform where the FFB placed. The camera used is a DFM 22 BUC03 Imaging Source CMOS camera with 744 x 480 pixel sensor, 76 maximum FFS, 6 m x 6 m pixel size. This camera is accompanied by software for recording images and saving in many file formats. This camera is a RGB camera which has a USB connection for data transfer convenience. The experiment was done by illuminating the samples using room light. The FFBs were placed on a platform which layered by white sheet to obtain white background. Images of front and back side of FFB are taken using the accompany software of the camera and saved in bitmap files for processing process. The distance from camera and FFB are adjusted hence the whole FFB image can be captured for small and big size FFB. The distances are then kept constant. Applied Science and Technology, Vol.1 No

4 3.3. Image Processing After the samples were illuminated, the oil palm FFB images were captured by the CMOS camera and saved in bitmap or jpeg format for further processing. ImageJ 1.47v software was used to get the RGB values, number of outer fruits, length and equatorial diameter, estimated density. There are 4 features in ImageJ software used for image processing. First is to find the intensity or value for RGB component using RGB plot feature, and the second is to find average RGB Intensity using (R+G+B)/3. Both features were performed using edge detection to reduce background. Third feature is to calculate length and equatorial diameter of the FFBs. The fourth is to use particle analyze feature in ImageJ to calculate the number of outer fruits of FFB. For the last two processes, the images were converted first to gray scale and two calibrations were performed before calculating the length and number of fruits. To measure the length and diameter, conversion from pixel to cm is needed. This was done using a ruler image on top of a FFB taken at the same distance as other images. For this process, a line is drawn from one edge to the other edge, from top end to basal section end for length) and from one end to the other end of equatorial section. The number of pixels counted is converted to cm. The second calibration is to set the size of fruit to be calculated, it was done using images with10-20 oil palm loose fruits located at some distances, the imagej cell counting plugin or particle analyze was set so that the calculation result is the same as the fruit number. Then the setting was applied to the FFB images. The conversion from pixel to cm obtained and used was pixels for 1 cm, the setting for fruit counting was pixels. 4. Results and Discussion Figure 1 shows the graphs of color component of FFB image for front and backside FFB of Tenera Type A. Red components more dominated and higher for front side for over ripe FFB (F4) since the FFB color is orange to dark red. For back side, the profile is almost flat except for F4. It is consistent with other researches where the red component for unripe and over ripe almost the same and slightly higher for ripe FFB [5,6,7]. Figure 1. Color components of RGB values for front and back side FFBs of Tenera A Figure 2 shows the graph of color component of FFB image for front and backside of Tenera B. Red component is also dominant but higher for back side. Tenera type A and B are different in size and fruit distribution. Tenera B has smaller fruit size but higher fruit distribution even for backside. Higher red could be because its color is lighter at back side. Figure 2. Color components of RGB values for front and back side FFBs of Tenera B Applied Science and Technology, Vol.1 No

5 Figure 3. Average RGB Intensity of Tenera A and Tenera B Figure 3 shows the average RGB intensity for both Tenera type. From Figure 3, it is shown that the intensity get higher as the ripeness increases for front side of FBB and are different for back side of FFB images. The variation of the RGB intensity comes from two sources, one is from the variation of FFB size for each sample and distribution of fruits on the FFBs. The second source is from applying the ImageJ process. There is also slightly irregularity in color if the whole FFB image is taken as region of interest (ROI), even the background around FFB has been subtracted, and each side of FFB has lighter color in some section especially at basal part due to frond mark. Figure 4 shows the thresholding results for counting the fruit numbers. It can be shown that the size of the fruits to be counted are not the same, this is due to thresholding adjustment between dark and white, and irregularity in texture of the FFB surfaces. Figure 4. Thresholding images for Number of Fruit Calculations for Tenera A and B Figure 5. Fruit Counting results for Tenera A and B Figure 5 shows the results of counting the number of outer oil palm fruits. It shows that the number of fruits decreases as the ripeness increases, however the variations are higher for overripe2 (F5) because 2/3 or the fruits have detached and varied for each FFB sample. Tenera A has irregularity because it has bigger fuit size, and bigger fruitlets. The percentage differences range from 1 15 % except for overripe FFB (F5) bigger tha 55 %. Applied Science and Technology, Vol.1 No

6 Comparison for measurement of length, equatorial length, and estimation of FFB density has also been done. The length of FFBs measured manually ranges from cm cm for Tenera A dan cm to cm for Tenera B. The equatorial diameters range from for Tenera A and cm for Tenera B. These results are for averaged for all samples. The results of imagesj are smaller, the length of FFBs ranges from cm for Tenera A and cm for Tenera B. The equatorial diameter ranges from cm for Tenera A and cm for Tenera B. The differences between two measurements due to curvature of the FFB where the length and diameter measured manually taken by taking circumference and dividing by two. The correction was measured. It was about 8-10 cm which means about 4-5 cm for each side of oil palm FFB. The density was estimated as the density of an ellipsoid with mass of the FFBs has been measured. The results ranges from kg/m2, the results should be bigger than the density of water, however the errors for measuring length and diameter were accumulated on the density estimation Conclusion This study was a preliminary study of applying ImageJ software to analyze the physical properties of FFBs and comparison to real manual measurement. There are some aspects needs to be addressed for future research due to the complexities of FFB structures and varieties. The comparison should be done also using soft computing programs. There are many pluggins of ImageJ software that have not been explored in this research such as 3D pluggins. One of weakness of imagej software is interfacing which is very important for automation system. This matter has been addressed by many researches lately. Some findings can be summarized from this study. The results show the average RGB intensities of Tenera A are slightly higher than Tenera B for all ripeness stages due to bigger fruit size. Percentage differences of physical properties measured by both methods range from 1% -15 % except for over ripe FFBs due to irregularity in fruit number and density. ImageJ software was able to used for image processing especially for those who lack of soft computing skill or does not have much time to do it. Acknowledgement. This research was supported by Universitas Riau and funded by DRPM under the Ministry of Research, Technology, and Higher Education (Hibah Bersaing Research Grant: 455/UN /LT/2016). The authors would like to thank Faculty of Agriculture Universitas Riau and CARL Plantations for providing the FFB samples. References [1] Sunilkumar, K. and Babu, D. S., Surface color based prediction of oil content in oil palm (Elaeis guineensis Jacq.) fresh fruit bunch, African Journal of Agricultural Research 8 (6), (2013). [2] Solovchenko, A.E., Chivkunova, O.B., Gitelson, A.A., Merzlyak, M.N., Fresh Produce: Non- Destructive estimation pigment content, ripening, quality, and damage in apple fruit with spectral reflectance in the visible range, Global Science Books, 2010.Athmaselvi, KA, Jenney, P., C. Pavithro, I. Ray. Physical and biochemical properties of selected tropical fruits, Int. Agrophys., 28, (2014). [3] Razali, M.H, Somad, A.S, Halim, M.A, Roslan, S., A Review On Crop Plant Production And Ripeness Forecasting, International Journal of Agriculture and Crop Sciences, 4 (2), (2012). [4] Randhawa, H. S. and Sharma S., A Survey of Computer Vision and Soft Computing Techniques for Ripeness Grading of Fruits, Journal of Advanced Computing and Communication Technologies 2 (4) (2014). [5] Fadilah, N. dan J. Mohamad-Saleh, Z.A. Halim, H. Ibrahim, S.S. Syed Ali Intelligent ColorVision System for Ripeness Clasification of Oil Palm Fresh Fruit Bunch. Sensor(Basel) Vol 12 (10): (2012). [6] Alfatni, M.S.M.; Shariff, A.R.M.; Shafri, H.Z.M.; Saaed, O.M.B.; Eshanta, O.M., Oil palm fruit bunch grading system using red, green and blue digital number, J. Appl. Sci. 8, (2008). [7] Ghazali, K.H.; Samad, R.; Arshad, N.W.; Karim, R.A., Image Processing Analysis of Oil Palm Fruits for Automatic Grading, In Proceedings of the International Conference on Instrumentation, Control & Automation, (2009). [8] Collins, T.J, ImageJ for microscopy, BioTechniques 43:S25-S30 (2007). Applied Science and Technology, Vol.1 No

7 [9] Peressotti, E., Duchêne, E., Merdinoglu, D., and Mestre, P., A semi-automatic non-destructive method to quantify downy mildew sporulation, Journal of Microbiological Methods, 84, (2011). [10] Girish V. and Vijayalakshmi, A., Affordable image analysis using NIH ImageJ. Indian J. Cancer 41:47 (2004). [11] Lino, A. C. L., Sanches, J., Dal Fabbro, I. M., "Image processing techniques for lemons and tomatoes classification," Bragantia, 67(3) (2008). [12] Ismail, W.I.W.; Bardaie, M.Z.; Hamid, A.M.A., Optical properties for mechanical harvesting of oil palm FFB, J. Oil Palm Res. 12, (2000). [13] BACP, Practical Guides: Budidaya Kelapa Sawit Ramah Lingkungan untuk Petani Kecil, BACP - PanEco Booklet PP, 2014, Diakses pada tanggal 20 September [14] Roseleena, J., J. Nursuriati, J. A.hmed, dan C.Y. Low.. Assessment of palm oil fresh fruit bunches using photogrammetric grading system, International Food Research Journal 18(3): (2011). [15] Makky, M. Soni, P. Salokhe, Vi. M., Automatic non-destructive quality inspection system for oil palm fruits, Int. Agrophys., 28, (2014). Applied Science and Technology, Vol.1 No