Ripeness detection of oil palm fresh fruit bunches using fluorescence sensor
Classification of oil palm fresh fruit bunches (FFB) into its correct ripeness category (under-ripe, ripe and over-ripe) is a critical factor that dictates efficient oil palm milling operations. This study investigates the fluorescence sensor to determine which excitation LEDs are suitable in discr...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/41818/1/FK%202011%20151R.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Classification of oil palm fresh fruit bunches (FFB) into its correct ripeness category (under-ripe, ripe and over-ripe) is a critical factor that dictates efficient oil palm milling operations. This study investigates the fluorescence sensor to determine which excitation
LEDs are suitable in discriminating between the different ripeness categories.
To determine this, we used a Multiplex®3 sensor, which has an active fluorescence sensor system; comprising of 9 excitation LED (6 UV and 3 RGB for red, green and blue excitations) and 3 photodiodes for the emitted fluorescence in the yellow (590 nm),red (685 nm) and far red ranges (735 nm), to detect oil palm FFB ripeness categories. The in-field signal value data were collected using the sensor system from a total of one hundred and eighty (180) oil palm FFB. These oil palms FFB were classified into underripe,ripe, and over-ripe ripeness categories.
Feature selection method, the rank method based on chi-square value was used to select the best predictors among available features. Fourteen classification methods (SPSS
Classification TreeTM (QUEST, CHAID, and CRT), SPSS Discriminant Analysis (Enter independent together), STATISTICA Stochastic Gradient Boosting Trees, STATISTICA
Interactive Tree (C&RT), STATISTICA MARSplines, STATISTICA General Stepwise Linear Discriminant Analysis, STATISTICA Automated Neural Networks Classification, STATISTICA Random Forest For Classification, Machine Learning (Support Vectors Machine, Naïve Bayes Classifier and k-Nearest Neighbour)),were used to assess the applicability of using the sensor system. Based on the classification accuracies, data analysis on the predictors indicated that the signal values of the data could be valuable in predicting the maturity stage of the oil palm FFB. The STATISTICA Stochastic Gradient Boosting Trees yielded highest average overall accuracies of 89.4% for the correct classification of oil palm FFB using the blue to red fluorescence ratio (BRR_FRF) as a predictor. Additionally, the average individual classes (under-ripe, ripe and over-ripe) classification accuracies were also higher than
76%. Thus, fluorescence sensing using the blue to red fluorescence ratio (BRR_FRF) as a predictor is useful for oil palm FFB ripeness detection under field conditions. This research will be useful for future development of low cost non-destructive, automatic and real time oil palm FFB grading system. |
|---|