Computer vision and NIR spectroscopy: an intelligent solution to optimize the fresh fruit bunches quality assessment in the Colombian oil palm agroindustry

Achieving high palm oil efficiencies requires strict control of FFB (fresh fruit bunches) as well as efficient control during the oil extraction process. One of the main routine activities carried out in the mills is FFB quality assessment. Currently, FFB inspection is carried out visually by workers, considering previously established ripeness criteria. However, results obtained by worker inspection are subject to human bias, and changes in lighting conditions among other factors. Low sample representativeness per shipment (5%) is another major drawback of traditional human inspection. Thus, improvement opportunities in terms of subjective quality measurements and statistical representativeness have been identified. Cenipalma has been developing prototypes and automated qualification systems based on two fundamental pillars of Data Science: Computer Vision (artificial intelligence) and NIR spectroscopy (NIRS). In the Computer Vision pillar, algorithms based on image segmentation and classification have been developed. Simultaneously, based on NIR technology, prediction models are being developed for variables associated with the evaluation of oil content, moisture, and other conditions related to the FFB ripeness scale. Currently, there is already a prototype of this system under real-time training and improvement in one of the POMS in Colombia. Therefore, this innovation project aims to allow the establishment of dynamics of negotiation and continuous improvement on reliable criteria based on sufficient quantity and quality of information between plantations and POM.

Please consider downloading this presentation: https://shorturl.at/msxR9

This is a video of a prototype: https://shorturl.at/ADF45