Development of Convolutional Neural Network Model for Abaca Fiber Stripping Quality Classification System

Grading the quality of abaca fiber is employed for its efficient and different uses in the global market. Recent studies focused on grading the abaca fiber based on its normal grade. In this study, we introduced a device that classifies abaca fiber based on it stripping quality. We applied Convolutional Neural Network (CNN) VGGNet-16 architecture to train and develop a model. In model training, we gathered 300 images of excellent fiber, 300 for good fiber, and 200 training images of a fiber with fair quality. The model training employed the 80/20 ratio wherein out of the 800 training images of gathered Abaca fiber, 80% are used as training data and the 20% are utilized to validate the efficiency and accuracy of the model. The result also shows an overall classification accuracy of 96.7 % upon its testing. In addition, to create a working prototype, we use raspberry pi 4 as the microcontroller of the device which integrated both the hardware and the software components. Although, very satisfactory classification accuracy was obtained, the model can be applied to continual learning scenario to see whether the catastrophic forgetting phenomenon will also occur in this kind of tasks.