Trash detection in an aquatic environment

Abstract

Trash in the water bodies is alarming for the world nowadays. As it is a limited source, pollution caused by the trash is threatening for the environment, wild-life and long term effect for the human health and economy. In developing countries like Bangladesh, where a large number of the species of biodiversity belongs to the aquatic environment, pollution caused by waste can have serious consequences for the economy and the endangered species of the biodiversity. In order to address these issues, this paper presents the analysis and performances of the detection models which will be beneficial in the future to implement an aquatic environment waste detection system. By utilizing deep learning techniques the system will be able to analyze data from edge devices and make accurate predictions about the presence of trash under the water. After reviewing a plenty of papers, we have noticed that the other detection models such as YOLOv2, Inceptionv3, Mask-R CNN and CNN require a lot of time and provide less accuracy compared to YOLOv5 and YOLOv7. That’s why we have chosen these algorithm models to analyze our dataset. We have also tried to implement a new algorithm called EfficientDet which is an object detection algorithm that combines efficiency and accuracy. It was introduced by Google in 2019. In this paper, we have prepared our self-prepared dataset which includes 1400+ pieces of data collected from various sources like St. Martin’s Island, pond, roadside drains and so on. Then we have processed and train the data and run three algorithm models which includes YOLOv5 , YOLOv7 and EfficientDet and got accuracy of 97%, 93% and 96%. The challenges of our work were to detect the trashes in the polluted water where the light is not sufficient. Because in the deep of the sea or any water source where the light is minimal, the detection of trashes become difficult. We hope to enrich our dataset more in the future and aim to build a model using raspberry pi or arduino to use the progressive algorithm models by eradicating the challenges of underwater trash detection.