Oil Palm Ripeness Detector (OPRID) and Non-Destructive Thermal Method of Palm Oil Quality Estimation

Transcription

1 Oil Palm Ripeness Detector (OPRID) and Non-Destructive Thermal Method of Palm Oil Quality Estimation Abdul Rashid Mohamed Shariff, Shahrzad Zolfagharnassab, Alhadi Aiad H. Ben Dayaf, Goh Jia Quan, Adel Tursun,, Assoc. Prof. Rimfiel Janius, Assoc Prof. Dr.Hawa ZE Jaafar, Prof.Dr.Aris Ishak, Prof.Dr.Reza Ehsani

2 INTRODUCTION Oil palm (Elaeis guineensis) was first introduced to Malaysia as an ornamental plant in Since 1960, planted area had increased at a rapid pace. In 1985, 1.5 million hectares were planted with palm tree, and it had increased to 4.3 million hectares in It has become the most important commodity crop in Malaysia. As of 2011, the total planted area was million hectares.(mpob 2011)

3 INTRODUCTION Oil palm is the most productive vegetable oil crop, capable of producing 4.27 t of palm oil per hectare per year.(mpob 2011) Malaysia with producing up to 18,400 thousand metric tons is the second largest producer of palm oil and %86 of worldwide export.(usda, 2011).

4 First (1 st ) Part Of The Presentation: FFBs maturity classification and oil analysis correlation

5 . INTRODUCTION Grading of oil palm fruits is conventionally observed by human vision. This is used for ripeness classification. Today many types of research has been carried out to find the correlation between oil content and quality of the oil palm fruit based on its colour (RGB). These researches followed more advanced methods and techniques by using different types of sensing techniques. Accordingly some near sensing devices has been constructed in order to achieve real time classification result of certain fruit. Oil Palm Fresh Fruit Bunches Ripeness Detector (OPRID) using multi spectral bands has been designed by integrated different sensors and sources of illumination.

6 Problem Statement A fast, accurate, and objective ripeness classification of oil palm fruit bunches for unripe, under-ripe, ripe, and over-ripe grading in real time has its limitations using traditional methods. Traditional methods of oil palm FFB quality assessment are costly and tedious. There is a need not only for automatic detection of ripeness but also automatic determination of oil content and oil quality parameters

7 Objectives The main objectives of this work are: To determine maturity classification of FFB using OPRID. To determine correlation between OPRID signals and oil palm parameters.

8 Technical Information Of OPRID

9 OPRID calibration Sensors calibration. OPRID preparation steps. Sensors require calibration to quantify the sensor s response to known radiometric input and to characterize the interactions and dependencies between the optical, mechanical, and electronic components Mode 1 White balance calibration: using pure and smooth white surface with area that fit OPRID slot, then record all AU readings using four detectors with ten LEDs separately (40 models). Mode 2 Black level calibration: using same procedure in mode1 but change white surface with black one.

10 LEDs Adjusting OPRID preparation steps. For using the device as efficiently as possible, where OPRID sensors have high spectral resolution (AU=65280) and LEDs power can be adjusted in many levels, so LEDs should be adjusted to achieve highest performance of device, and this will be done but with two consideration: First, avoid reaching saturation point, where saturation means that the use of this reading cannot be relied upon in the classification tasks and comparison between certain values. Secondly, the LED power should be stronger enough to extent AUs range that makes the task of distinguishing between readings is more effective.

11 Research flowchart 1. Data collection 2. Data preprocessing 3. Data analysis

12 Methodology

13 Methodology Eliminated

14 Methodology

15 Methodology

16 Results (1 st experiment of FFB)

17 Results (2 nd experiment of FFB)

18 Results FFB Experiment Classification Accuracy ,5 98,5 98,5 98,5 98,5 98,5 98,5 98, st exp. 2nd exp ,5 95,5 95,5 95,5 95, Green S1 Red S2 DRedS2 FRedS3 BlueS4 GreenS4 AmberS4 RedS4 DRedS4 FRedS4 IRedS4

19 Results Highest accuracy Models Algorithms First experiment 100% RedS2, RedS4, FRedS4, IRedS4 Logistic, Simple Logistic, LMT, Second experiment 98.5% RedS4 Logistic

20 Summary of Results Able to determine ripeness of FFB by classifying them into four categories of unripe, under ripe, ripe and over ripe. Able to check classification accuracy of FFB by achieving accuracy of 100%. Also able to determine the best algorithms for classification of FFB which is the LMT and RandomForest.

21 Oil Analysis

22 Oil Analysis (2)

23 Correlation between OPRID and laboratory classification

24 Second (2 nd ) Part Of The Presentation Oil Palm quality parameter estimation based on thermal imaging technique

25 INTRODUCTION Temperature measurement is an important aspect in any industrial process and infrared thermography has revolutionized the concept of temperature measurement ( R.Vadivambal, 2010). Temperature measurements mostly performed using some contact instrument like thermometers, thermocouples, thermistors, and resistance temperature detectors. While infrared thermal imaging is a non-contact, nondestructive technique which provides temperature mapping of the material( R.Vadivambal, 2010). Therefore, use of infrared thermal imaging is widely increasing in many fields.

26 Problem statement The maturity or ripeness of the oil palm fruits influences the quality of oil palm. Conventional method includes manual detection of FFB ripeness by counting the number of loosened fruits per bunch. This manual sorting of oil palm FFB is a time-consuming, costly, needs many workers and the results may have the human error. Currently, grading of palm oil fruit is performed through visual inspection using the surface color as the main quality attribute. Color is the most important indication of FFB ripeness but there is no inspection of the relationship between color and optimum FFB oil content, FFA, PV, DOBI and CAROTIN

27 Objectives This research investigates the potential of infrared images (Thermal images) as a predictor to estimate the oil content, Free Fatty Acid (FFA), peroxide value (PV), Deterioration of Bleachability Index (DOBI) and Carotene.. The research objective are: To investigate the correlation of the thermal image oil content, Free Fatty Acid (FFA), peroxide value (PV), Deterioration of Bleachability Index (DOBI) and Carotene. To develope a technique to predict the total pecentage of oil content and Free Fatty Acid (FFA), peroxide value (PV), Deterioration of Bleachability Index (DOBI) and Carotene in the fresh fruit bunch.

28 METHODOLOGY Obtaining raw data from the thermal camera is a temperature array of the thermal distribution of the object s surface. Transforming this temperature array into an image format to create a thermal image and extract the relevant feature polygon of FFB by using thermal image processing software FLIR reporter wizard and thermacam researcher pro 2.10 Determine the relationship between oil content, Free Fatty Acid (FFA), peroxide value (PV), Deterioration of Bleachability Index (DOBI) and Carotene in the FFB and temperature data from the thermal images by using the correlation and regression techniques. Development a predictive model using Artificial Neural network (ANN) to estimate the total percentage of oil content, FFA and other quality parameters

29 Fresh fruit bunches of varying ripeness based on the number of loosened fruit and visual observation of fruit color were harvested from United Plantation Research and Development (UPRD) Center in Teluk Intan, Perak Malaysia. 135 harvested bunches were weighted and thermal images from three sides of the fresh fruit bunches were then captured with a FLIR E60 camera The oil palm bunches were from the Nigresens cultivar according to three maturity categories: Under Ripe, Ripe and Over Ripe. For each category, the images were collected based on three types of weight: 0-15 kg, 15-25kg and above 25kg. First Data collection

30 After finishing the capturing session for the FFB image, the bunches will be send to laboratory for chemical analysis to determine its oil content, FFA, PV, DOBI and carotene.

31 First Data Analysis

32

33 The data were collected with two thermal camera E60 and T440. Second Data Collection The two set of data from three sides of 135 FFB were collected in United Plantation Research and Development in Teluk Intan(UPRD), Perak. The oil palm bunches same as first data were from the Nigresens cultivar according to three maturity categories: Under Ripe, Ripe and Over Ripe.

34

35 Image processing a:under Ripe FFB b: Ripe FFB c: Over Ripe FFB

36 Normality and Homogeneity Test

37 Temperature and FFA Temperature = Palm temperature Atmosphere temperature

38 Temperature and Oil Content

39 Temperature and PV 0,4 0,35 0,3 0,25 0,2 0,15 0,1 0,37 0,32 0,27 0,

40 Temperature and DOBI 4,4 4,2 4 3,8 3,6 4,3 4,2 3,4 3, ,6

41 Temperature and Oil Content ,5 26,62 26, , ,07 24,5 24 UR R OR

42 Temperature and FFA 2

43 Correlation coefficient P= and r= There is a strong relationship between temperature and FFA

44 Temperature and DOBI 2

45 Temperature and CAROTINE 2

46 Oil content prediction By ANN Each group has 90 samples

47 Principal Component analysis

48 PCA: Extraction Method

49 Scatterplot Matrix of Component Scores

50 Oil content prediction By ANN with reduce feature based on PCA

51 FFA prediction by PCA-ANN Each group has 78 samples

52 PV prediction by PCA-ANN Each group has 40 samples

53 DOBI prediction by PCA-ANN

54 Summary of Results The performance of the ANN for Oil content, FFA and other quality parameters prediction of oil palm FFB was investigated using two methods: training ANN with full features and training ANN with reduced features based on the Principal Component Analysis (PCA) data reduction technique. PCA-ANN predictor can be good indicator to predict oil content and oil quality parameters: Oil content 94.6% FFA 77% PV 83.3% DOBI 75.9%

55 Conclusion This technique offers a non-destructive means of assessing palm oil quality and can enable oil yield and/or oil quality (FFA) determination. This study allows for rapid screening of FFB ripeness and relates it to oil content in the FFB and accordingly the amount of oil that can be extracted from a consignment of FFB arriving at the mill

56 GRACIAS THANK YOU TERIMA-KASIH