Integrated deep and classical machine learning for ripeness classification of oil-palm fresh fruit bunches

The palm oil industry plays a significant role in the global market, driving continuous innovation and improvements in production efficiency. A critical stage in the palm oil production process is the sorting of fresh fruit bunches (FFB) at the loading ramp reception area, which directly impacts product quality. However, the current manual inspection methods rely on single-visual assessments, often leading to inconsistencies and misclassification of FFB ripeness. Inaccurate classification, particularly of unripe FFB, can affect both the yield and quality of crude palm oil (CPO). This study introduces the application of computer vision technology to enhance the FFB sorting process. By leveraging distinct visual characteristics such as colour and the presence of detached fruits, the study integrates deep learning (CNN) and machine learning (HOG-SVM) techniques for improved classification accuracy. The proposed system achieves an overall accuracy of 90% with an average processing time of approximately 30 seconds, demonstrating a significant enhancement over traditional manual sorting methods. This advancement offers a more efficient and reliable approach to FFB inspection, ultimately contributing to improved processing efficiency and product quality.