Detection of garbage outside of the deposition equipment: A study on classification-based and object detection-based computer vision approaches

• Autor(a) principal: Soares, Gonçalo Filipe Constantino
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/30320
• topic: Redes neuronais -- Neural networks; Waste; Visão computacional -- Computer vision; Identification; Garbage; Machine learning; Resíduos; Identificação; Lixo

Excessive waste accumulation is a problem in large cities and capitals where urban waste production is high. This problem leads waste collection teams to make a greater effort to combat the situation. Therefore, this dissertation proposes two waste identification systems, which will be compared, to solve this problem in Portugal’s capital. The main objective of this proposal is to facilitate the work of waste collection in the city of Lisbon, which is carried out by the teams of the Lisbon Waste Collection Centers. In order to facilitate and help waste collection, Lisbon City Council collaborated with the collection team inspectors and created ”LxDataLab”, a platform that provides a variety of datasets. The images are taken from smartphone cameras by the collection crews and are usually taken from moving vehicles or even local residents. Image processing is carried out differently in the two systems created. The patch-based garbage detector system uses the original images, changes their resolution and splits the image into several sub-images that contain a portion of the altered image. This system uses hand-made, pre-trained neural networks and different methods to obtain the results using the dataset of sub- images. The other system, called the object detection-based system, uses the original images and evaluates them using an algorithm called yolov5. Finally, a fair comparison is made between the two systems to determine which is the most effective by evaluating the accuracy and loss values.