[Paper Review] Sparse Coral Classification Using Deep Convolutional Neural Networks

Mohamed Elawady|arXiv (Cornell University)|Nov 29, 2015

Coral and Marine Ecosystems Studies4 references22 citations

TL;DR

This paper proposes a sparse deep learning framework using Convolutional Neural Networks (CNNs) for automatic coral species classification in underwater images, integrating texture and shape features (WLD, PC, ZCA) with color channels and enhancing images via preprocessing. It achieves improved classification accuracy on two novel datasets—Moorea Labeled Corals and the new Atlantic Deep Sea coral dataset—marking the first application of CNNs to underwater coral image classification.

ABSTRACT

Autonomous repair of deep-sea coral reefs is a recent proposed idea to support the oceans ecosystem in which is vital for commercial fishing, tourism and other species. This idea can be operated through using many small autonomous underwater vehicles (AUVs) and swarm intelligence techniques to locate and replace chunks of coral which have been broken off, thus enabling re-growth and maintaining the habitat. The aim of this project is developing machine vision algorithms to enable an underwater robot to locate a coral reef and a chunk of coral on the seabed and prompt the robot to pick it up. Although there is no literature on this particular problem, related work on fish counting may give some insight into the problem. The technical challenges are principally due to the potential lack of clarity of the water and platform stabilization as well as spurious artifacts (rocks, fish, and crabs). We present an efficient sparse classification for coral species using supervised deep learning method called Convolutional Neural Networks (CNNs). We compute Weber Local Descriptor (WLD), Phase Congruency (PC), and Zero Component Analysis (ZCA) Whitening to extract shape and texture feature descriptors, which are employed to be supplementary channels (feature-based maps) besides basic spatial color channels (spatial-based maps) of coral input image, we also experiment state-of-art preprocessing underwater algorithms for image enhancement and color normalization and color conversion adjustment. Our proposed coral classification method is developed under MATLAB platform, and evaluated by two different coral datasets (University of California San Diego's Moorea Labeled Corals, and Heriot-Watt University's Atlantic Deep Sea).

Motivation & Objective

• Develop an autonomous underwater system for coral reef repair using AUVs and machine vision.
• Address the challenge of classifying coral species in low-visibility, noisy underwater images with artifacts like rocks and marine life.
• Create a robust, supervised deep learning framework for sparse coral classification in deep-sea and shallow-sea environments.
• Introduce a new labeled dataset of cold-water corals from the Atlantic deep sea to support underwater coral research.
• Improve classification performance through multi-channel feature fusion and advanced image preprocessing techniques.

Proposed method

• Employ a CNN architecture with hybrid patching to handle variable-sized image regions around labeled coral points.
• Integrate supplementary feature maps (WLD, PC, ZCA-whitened) alongside spatial color channels to enrich input representation.
• Apply state-of-the-art underwater image enhancement techniques, including color normalization and histogram equalization.
• Use min-max and z-score normalization to standardize input image data across datasets.
• Train the CNN using backpropagation with early stopping to prevent overfitting, particularly on the MLC dataset.
• Optimize network architecture through ablation studies on hidden layer capacity, feature maps, and normalization methods.

Experimental results

Research questions

• RQ1 Can deep convolutional neural networks effectively classify coral species in challenging underwater imaging conditions?
• RQ2 How do supplementary texture and shape features (WLD, PC, ZCA) improve classification accuracy compared to color-only inputs?
• RQ3 What is the impact of image preprocessing (color normalization, enhancement) on coral classification performance?
• RQ4 How does hybrid patching and multi-channel input data affect model generalization across different coral types and depths?
• RQ5 What are the optimal network hyperparameters and architecture configurations for sparse coral classification in underwater settings?

Key findings

• The proposed CNN with multi-channel input (color + texture/shape features) achieved higher classification accuracy than baseline color-only models.
• Image preprocessing, especially color normalization and enhancement, significantly improved performance on both the Moorea Labeled Corals and Atlantic Deep Sea datasets.
• Overfitting was observed in the MLC dataset after epoch 30, indicating the need for regularization or architectural constraints.
• Training and test errors plateaued after epoch 35 in the ADS dataset, suggesting convergence with current learning rate and architecture.
• The hybrid patching strategy enabled effective multi-scale feature extraction from variable-sized coral regions.
• The introduction of the Atlantic Deep Sea dataset provides a new benchmark for cold-water coral classification, supporting future underwater robotics research.