Development of Natural Fiber Classification System using Deep Convolutional Neural Network

This study developed a prototype for classification of natural fibers in the Philippines using Convolutional Neural Network.The study utilized AlexNet CNN architecture for its model development, OpenCV and Flask framework for user-interface development, and Raspberry Pi as a major component for hardware development.The natural fibers included in the study were Abaca, Coconut, Pineapple, Raffia, and Salago fiber.Each class has a 200 training images wherein it applied the 80:20 ratio, in which the 20% was used for validation.Using the developed prototype, the researchers also conducted a system testing.During the testing, 30 samples were used for each class.Upon testing, the model achieved an accuracy rate of 94% wherein out of 30 samples, the system correctly predicted all the samples for both Coconut and Salago fibers, 28 for Abaca and Pineapple fibers, and 25 for Raffia fiber.Overall, the developed system can be utilized as an alternative device for classification of natural fibers.