Enhancement of Murky Underwater Images for Optimum Object Detection

Abstract

Underwater object detection has been a continuous challenge due to its unpredictable murkiness. Murkiness of water are caused by the scattering of lights, weather condition as well as the growth of algae. Loss in visibility due to murkiness made it harder to do object detection underwater. This research project aims to enhance object detection in murky underwater images through image enhancement techniques. The project consists of three main stages: collecting and categorizing a dataset of murky underwater images, applying image enhancement methods, and implementing the You Only Look Once version 5s (YOLOv5s) object detection algorithm for accuracy comparison. The dataset includes separate sets of clear and murky images for training, validation, and testing. Image enhancement techniques, such as Contrast Limited Adaptive Histogram Equalization (CLAHE), grayscale, and colour correction, were utilized to improve the clarity of the murky images. Evaluation was conducted using the Peak Signal-to-Noise Ratio (PSNR) metric. The results show that the CLAHE and grayscale technique improved object detection accuracy by 10% compared to the original images. These findings have significant implications for search and rescue operations and marine conservation efforts.