Palm Fresh Fruit Bunches (FFBs) Colour Grading System Using Raspberry Pi

Abstract

In Malaysia’s palm oil industries, the classification of the palm fresh fruit bunches (FFBs) is still manually done by human inspection. Human error is the most common thing to happen in the process of classification. Though, humans can be trained to be perfect yet the classification of FFBs is a time-consuming process. This report presents Colour Grading System Using Raspberry Pi. In this study, data collection was obtained from the ROBOFLOW database, it comes with 3 different categories, Ripe, Unripe and Overripe. The method uses Convolutional Neural Networks with untrained ResNet-50 architecture for deep learning layers and runs in MATLAB with the additional toolbox extension. Transfer Learning is a method to design a neural network by uploading the dataset of training and validating FFBs images. Hardware and software configurations were made to perform colour grading of FFBs using 2 camera angles. The result of the colour grading system shows that FFBs can be determined using MATLAB and Raspberry Pi. Other than that, LED lights also were used to indicate the ripeness of FFBs. The developed system manages to classify FFBs with 92% - 96% accuracy. Unfortunately, MATLAB does not support multiple cameras because the multi-camera board uses multiplexing circuits.