Object Recognition System with Image Enhancement for Lake Underwater Images

Abstract

The project aims to develop a system for recognizing objects in underwater images from lakes using a combination of image enhancement techniques and object recognition methods. A system is needed for underwater robots to distinguish objects in a lake while searching for them. Due to limited visibility in underwater photos, it is difficult to accurately capture the shape or color of objects. The objects in the database are chosen based on the assumption that they may fall into the lake and need to be found. Recognition methods for these objects are important for the underwater searching process. The techniques used are the CLAHE for contrast improvement and the YOLOv3 for object detection to improve the visual appearance and accurately identify objects. The proposed object recognition system uses Google Colaboratory as the development environment and the Python programming language to implement the system. The labeling process is done by using the LabelImg software. This project proposed the CLAHE_YOLOv3 method to overcome the problems. In this research, an image database for underwater images of lakes in various situations has been created. The developed database can be accessed at the following link: https://bit.ly/3klN9ex. The study compares the performance of the CLAHE_YOLOv3 method to the YOLOv3 technique for object recognition in underwater images. In the experiment, the setting is constant for depth from the surface which is 10 cm from the surface and the distance of the object from the camera is 25 cm. The CLAHE technique has improved and enhanced the local contrast in the image, making it appear sharper and the details clearer. In the experiment, the accuracy increases from 93.52% for YOLOv3 to 100% for CLAHE_YOLOv3.