[Paper Review] Algae Detection Using Computer Vision and Deep Learning

Arabinda Samantaray, Baijian Yang+2 more|arXiv (Cornell University)|Nov 27, 2018

Water Quality Monitoring Technologies27 references21 citations

TL;DR

This paper proposes a real-time, low-cost computer vision system using deep learning for detecting harmful algal blooms (HABs) in diverse aquatic environments. The system, deployable on UAVs, USVs, and other mobile platforms, leverages state-of-the-art object detectors (Faster R-CNN, R-FCN, SSD) and achieves high accuracy and speed, with R-FCN demonstrating the best balance of robustness, precision, and real-time performance on complex, cluttered imagery.

ABSTRACT

A disconcerting ramification of water pollution caused by burgeoning populations, rapid industrialization and modernization of agriculture, has been the exponential increase in the incidence of algal growth across the globe. Harmful algal blooms (HABs) have devastated fisheries, contaminated drinking water and killed livestock, resulting in economic losses to the tune of millions of dollars. Therefore, it is important to constantly monitor water bodies and identify any algae build-up so that prompt action against its accumulation can be taken and the harmful consequences can be avoided. In this paper, we propose a computer vision system based on deep learning for algae monitoring. The proposed system is fast, accurate and cheap, and it can be installed on any robotic platforms such as USVs and UAVs for autonomous algae monitoring. The experimental results demonstrate that the proposed system can detect algae in distinct environments regardless of the underlying hardware with high accuracy and in real time.

Motivation & Objective

• To address the growing threat of harmful algal blooms (HABs) that damage ecosystems, contaminate water, and cause economic losses.
• To overcome limitations of traditional algae monitoring methods, which are labor-intensive, expensive, and hardware-dependent.
• To develop a low-cost, real-time, and robust computer vision system for detecting algae across diverse environmental conditions and camera platforms.
• To enable deployment on mobile robotic platforms such as UAVs and USVs for autonomous, continuous water body monitoring.
• To improve detection accuracy in complex scenes with background clutter (e.g., trees, buildings) using deep learning-based object detection.

Proposed method

• The system employs three state-of-the-art deep learning object detectors: Faster R-CNN, R-FCN, and SSD, for end-to-end algae detection and localization.
• A custom dataset of real-world images from ground and aerial views was collected, including complex scenes with significant background clutter such as vegetation, buildings, and roads.
• The models were trained and evaluated using the mean Average Precision (mAP) metric, with Intersection over Union (IoU) used to compute detection correctness.
• Detection performance was assessed using precision-recall curves and IoU thresholds (e.g., 0.5), where IoU = |Box_pred ∩ Box_gt| / |Box_pred ∪ Box_gt|.
• The system was tested on both validation and test sets, with the test set intentionally composed of the most complex images to evaluate robustness.
• Real-time performance was evaluated on both CPU and GPU to assess suitability for deployment on fast-moving platforms like UAVs and USVs.

Experimental results

Research questions

• RQ1 Can a deep learning-based computer vision system detect algal blooms accurately across diverse environmental conditions and camera viewpoints?
• RQ2 How do state-of-the-art object detectors (Faster R-CNN, R-FCN, SSD) compare in terms of detection accuracy and inference speed for algae monitoring?
• RQ3 Can the system maintain high performance in the presence of significant background clutter such as trees, buildings, and vegetation without specialized hardware?
• RQ4 Is the system fast enough for real-time deployment on mobile robotic platforms like UAVs and USVs?
• RQ5 What are the primary failure modes of the system, and how can they be mitigated through improved data and model design?

Key findings

• The R-FCN model achieved the highest detection accuracy among the three models, with the highest mAP on both the validation and test datasets.
• SSD demonstrated the fastest inference speed on both CPU and GPU, making it highly suitable for real-time applications on mobile platforms.
• Despite the high complexity of the test dataset—including significant background clutter—the system successfully detected algae in both ground and aerial views.
• The mAP on the test set was lower than on the validation set, indicating that the system faces greater challenges in the most complex, real-world scenarios.
• Incorrect detections occurred when vegetation was misclassified as algae, primarily due to the shared green color, highlighting the need for more discriminative features.
• The system showed robustness to variations in camera parameters, image orientation, and environmental lighting, enabling reliable deployment across diverse platforms.