Development of sorting system for oil palm in vitro shoots using machine vision approach

There is a promising future for palm oil cultivation because of tissue culture technology that leads to increased production rates as well as enhanced quality. In spite of high demand for palm oil, cheap mass-produced plantlets of oil palm have yet to reach an attractive price. The main reason is that the processing of oil palm tissue culture is still operated manually and is considered too labour-intensive. This affects its economies, especially in developed countries where wages are high. In addition, the plantlet is susceptible to contamination throughout this process. Since a large proportion of manual work is at the in vitro shoots stage, and no commercial automated system for this task presently exists, research will develop a sorting task as a step to reach a fully efficient machine vision system. This research targets the sorting task which is one of four tasks which have been suggested to achieve greater automation of oil palm tissue culture while in the in vitro shoots stages. The four are separation, classification, sorting and rooting. In this work, the type of sample involves oil palm tissue culture (OPTC) shoots obtained from the Malaysia Palm Oil Board (MPOB) tissue culture laboratory. They were acquired, recorded, and utilized throughout the system development. A manual adjusting method for image intensity is here suggested to enhance and sharpen the acquired frames, and the region of interest has been determined as well, which led to simplifying the segmentation and reduce the preprocessing time. Region-based features, namely area, centroid, aspect ratio, extent and two cropping points have been represented in the shape of OPTC in vitro shoots. By using k-means algorithm the extracted features have been evaluated. A smart object tracking algorithm (SOTA) has been proposed for detecting and identifying the shoot on the conveyor belt. Based on SOTA and classification task decision, a sorting algorithm that can acquire, recognize, and eject a shoot has been improved and tested in an offline mode. Furthermore, workable values of external variables and a customized sorter have been designed to test the system in real-time mode and to smooth the ejecting process. Ultimately, the sorting algorithm performance came to be evaluated by support vector machine algorithm. The performance of the sorting process shows that acquiring, detecting, tracking and sorting functions operate well. The result of K-means has proven the robustness of the selected features. The resulting error of offline tests of the sorting algorithm did not exceed 4.33 per cent. The machine vision system in real-time can eject abnormal shoots from the conveyor, with a limited overall error that could reach as high as 6.6 per cent in the worst case. Close results between the performance of the developed sorting algorithm and SVM algorithm demonstrate that it is satisfactory and efficient. Automating the sorting task was achieved under the main goal which is increasing the production rate and enhancing the quality. In addition, the automated sorting system reduced the overstaffing which achieves part of the economies required to make it of interest in the industry.

Preview Text ITMA 2014 15 IR.pdf Download (1MB) | Preview

Actions (login required)

View Item