Identification and Sorting of Waste using Artificial Intelligence Based on Convolutional Neural Network

Abstract

Indonesia is one of the biggest producers of waste in the world. Every day there is rubbish that is not thrown in the trash, those that have been thrown in its place but are still wrong in distinguishing the type of garbage. In general, one of the types of waste is based on its nature, some are easy to decompose (organic), and some are difficult to decompose (inorganic). This research introduces five organic and inorganic waste objects by seeing whether the system can read objects at every position in the background and the number of objects in 1 picture frame. This system uses Tensorflow with a sample of 100 images for each object. Each sample object is given an identifier according to the object’s name and type. The sample is given an identifier for the system to learn to recognize the patterns and shapes of objects from the data objects we take. The results showed that the average percentage of system accuracy in detecting objects reached 90%, with the highest data accuracy reaching 99% tested on the banana peel, leaves, grass, cardboard, styrofoam, broken glass, bottles, cans, and nails. Furthermore, the authors try to put two or more different objects in the same frame and produce an average percentage value of 90%, as well as testing two or more similar objects in one frame simultaneously and producing an average value of 90%. From the test results, grass and bottle objects are samples that do not have read errors in the system design. For testing two or more objects in one frame, the author puts two similar objects and objects of different types. The result is that the system can recognize objects in the frame.