Penggunaan Capsule Network dalam Pengelompokan Plastik Berdasarkan Resin Identification Code

Abstract

Plastic is one of the most widely used materials in the world due to its technically advantageous properties and low cost. According to an analysis conducted by the organization Plastics Europe in 2022, global plastic consumption has steadily increased over the past 30 years, reaching 390 million tons in 2021, and is projected to rise to 600 million tons by 2025. Understanding plastic recycling codes is crucial to reducing improper usage and enhancing the potential for effective recycling. Therefore, automatic identification of plastic code types can serve as a means to support more efficient plastic waste management. This study employs the Capsule Network algorithm to identify seven types of plastic resin codes. The dataset used to train the CapsNet model consists of 1,585 image samples, with each class containing approximately 200 images on average. The images are processed through a convolutional layer, primary capsule layer, digit capsule layers, and fully connected layer. The model is evaluated using a test set of 150 images across eight recycling code classes, achieving a weighted average of 92% for precision, recall, and F1-score respectively. This plastic classification system can assist plastic users in improving recycling practices.