A Survey on ML-Based Techniques to Estimate Coral Reef Cover: State of the Art and Challenges

EasyChair Preprint 8415

Abstract

Even the slightest change in the environment can impact a sensitive ecosystem on earth, and the Coral ecosystem can be stated as one of them. The diver-based approach for monitoring coral reefs is very rigorous and has limitations in covering the area. The surge in the development of small and low power intelligent hydrobots (AUVs and ROVs) has given rise to automating the visual monitoring of reefs. These vehicles are smart enough to convert the raw data into information such as bleaching, biodiversity, and coral cover related to the health of the reef. The paper surveys recent advances in techniques to estimate coral reef Cover and identifies future research challenges. The study in this article attempts to understand various Artificial Intelligence integrated methods used in coral classification and estimation of coral cover. The recent advances in the Convolutional Neural Networks (CNN) like Inception v3, ResNet, and DenseNet have been used successfully. Highly efficient results are obtained in terms of speed and accuracy on datasets like EILAT and RSMAS. CNN has also been used to engineer feature extraction and semantic segmentation, to classify pixels in a given image into classes of interest in a supervised method. These segmented classes are then used for the estimation of coral cover. However, the main challenge is to obtain the training dataset for these algorithms. We used the segmentation aspect in the traditional machine learning technique using the random forest to mitigate this challenge and compared it to deep learning methods. Though segmentation methods using CNN like Unet are very efficient, methods involving traditional machine learning have also fetched impressive results on smaller datasets and in comparatively lesser computational time. Thus, envisaging that traditional machine learning methods can as well be utilized without involving much of resources and time.

Keyphrases: Artificial Intelligence, Coral Cover Estimation, Random Forest, autonomous vehicle, machine learning

@booklet{EasyChair:8415,
  author    = {Tessy Thomas and Pramod Maurya and B Manikandan and Nilesh Fal Dessai},
  title     = {A Survey on ML-Based Techniques to Estimate Coral Reef Cover: State of the Art and Challenges},
  howpublished = {EasyChair Preprint 8415},
  year      = {EasyChair, 2022}}