Computer Vision Methods For Coral Reef Assessment


The world’s natural systems are at a tipping point and many biologists believe that we are facing a massive ecological extinction event as a result of anthropogenic impacts on the planet (Pimm et al. 1995, Wilson 2002). Globally coral reefs face an unprecedented convergence of stressors including overfishing, pollution, diseases, destructive fishing practices, sedimentation, and coastal development that have led to led to massive global declines (Kline et al 2005, Jackson et al 2001, Hughes et al., 2003 ;Pandolfi et al 2005; Hoegh-Guldberg, 2007).


Traditional coral reef monitoring has required trained experts scuba diving on reefs to specify coverage and health of the key ecological groups (corals, algae, other invertebrates, fish, bare substrate, etc.). Recently, photographs or videos now routinely complement such surveys by experts but there is lacking a rapid, objective, quantitative, and automated classification of digital imagery. New technologies are needed to improve the efficiency and objectivity of surveys to assess the health of global coral reef communities on appropriate temporal and spatial scales. Analyses of these growing digital image archives remain extremely constrained by the extensive effort required by coral experts.

A convergence of several rapidly advancing technologies, including digital imaging, computational mass storage and processing speed, integrated with computer vision image analysis, now makes it feasible to acquire, archive, and digitally classify important aspects of coral reef community ecology and physiology. Computer vision technology has considerable potential to address these problems for coral reef ecosystems, but additional innovations by an interdisciplinary team are required to overcome challenges before a robust, automated cyber-enabled image analysis system can be confidently used for objective coral reef monitoring.

Main Challenges

  1. Developing computer vision methods to maximize the information that can be obtained from underwater digital images of coral reefs. The computer vision team will develop original segmentation and classification algorithms using our unprecedentedly large, unique data set of controlled, high resolution digital images of hundreds of tagged coral reef targets acquired over a 6 year period in Panama, and a 4 year period on the Great Barrier Reef. Portions of the datasets will first be classified by the coral ecology experts and then the classified images will be used to train the computer vision system, resulting in a state-of-the-art computational method for analyzing reef ecology.
  2. Development of an advanced underwater imaging system with co-registered RGB and fluorescence image planes to provide additional optical and physiological information that will increase the accuracy and speed of the classification process. Multispectral images, co-registered fluorescence images, turbidity data, ranging data, and other optical water column corrections and physiological measurements will be incorporated in order to maximize the discriminatory ability of the system. This system will be deployed as part of the continuing time-series in Panama and the Great Barrier Reef.