Aerial-trained deep learning networks for surveying cetaceans from satellite imagery

Borowicz, Alex; Le, Hieu Van; Humphries, Grant R. W.; Nehls, Georg; Höschle, Caroline; Kosarev, Vladislav; Lynch, Heather J.

doi:10.1371/journal.pone.0212532

Cited by 52 publications

(60 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Despite the promise that VHR satellites hold, currently images must be manually analysed by experts, which hinders their application in practice. While researchers would benefit from surveying large areas (e.g., searching for whales in open ocean [4]), manually scanning through enormous volumes of imagery becomes a near impossible task, particularly if surveys are to be repeated regularly [11]. Due to this most research has been limited to comparatively small scale studies over a few square kilometers [2].…”

Section: Introductionmentioning

confidence: 99%

Using Deep Learning to Count Albatrosses from Space: Assessing Results in Light of Ground Truth Uncertainty

et al. 2020

View full text Add to dashboard Cite

Many wildlife species inhabit inaccessible environments, limiting researchers ability to conduct essential population surveys. Recently, very high resolution (sub-metre) satellite imagery has enabled remote monitoring of certain species directly from space; however, manual analysis of the imagery is time-consuming, expensive and subjective. State-of-the-art deep learning approaches can automate this process; however, often image datasets are small, and uncertainty in ground truth labels can affect supervised training schemes and the interpretation of errors. In this paper, we investigate these challenges by conducting both manual and automated counts of nesting Wandering Albatrosses on four separate islands, captured by the 31 cm resolution WorldView-3 sensor. We collect counts from six observers, and train a convolutional neural network (U-Net) using leave-one-island-out cross-validation and different combinations of ground truth labels. We show that (1) interobserver variation in manual counts is significant and differs between the four islands, (2) the small dataset can limit the networks ability to generalise to unseen imagery and (3) the choice of ground truth labels can have a significant impact on our assessment of network performance. Our final results show the network detects albatrosses as accurately as human observers for two of the islands, while in the other two misclassifications are largely caused by the presence of noise, cloud cover and habitat, which was not present in the training dataset. While the results show promise, we stress the importance of considering these factors for any study where data is limited and observer confidence is variable.

show abstract

Section: Introductionmentioning

confidence: 99%

Using Deep Learning to Count Albatrosses from Space: Assessing Results in Light of Ground Truth Uncertainty

et al. 2020

View full text Add to dashboard Cite

show abstract

“…However, trained algorithms with known, quantifiable uncertainties may provide a more analytically uniform means of scanning, identifying and classifying FOIs. Studies 55,56 have shown remarkable progress in this field over recent years,however ongoing development and testing is still required to hone these methods in order to assess their accuracy when compared to visual observations in challenging conditions.…”

Section: Discussionmentioning

confidence: 99%

A comparison of baleen whale density estimates derived from overlapping satellite imagery and a shipborne survey

Bamford

Kelly

Rosa

et al. 2020

Sci Rep

View full text Add to dashboard Cite

As whales recover from commercial exploitation, they are increasing in abundance in habitats that they have been absent from for decades. However, studying the recovery and habitat use patterns of whales, particularly in remote and inaccessible regions, frequently poses logistical and economic challenges. Here we trial a new approach for measuring whale density in a remote area, using Very-High-Resolution WorldView-3 satellite imagery. This approach has capacity to provide sightings data to complement and assist traditional sightings surveys. We compare at-sea whale density estimates to estimates derived from satellite imagery collected at a similar time, and use suction-cup archival logger data to make an adjustment for surface availability. We demonstrate that satellite imagery can provide useful data on whale occurrence and density. Densities, when unadjusted for surface availability are shown to be considerably lower than those estimated by the ship survey. However, adjusted for surface availability and weather conditions (0.13 whales per km 2 , CV = 0.38), they fall within an order of magnitude of those derived by traditional line-transect estimates (0.33 whales per km 2 , CV = 0.09). Satellite surveys represent an exciting development for high-resolution image-based cetacean observation at sea, particularly in inaccessible regions, presenting opportunities for ongoing and future research. Gathering data on cetacean distribution and densities has traditionally employed visual observers operating from various platforms, typically either ships, aircraft or land 1-5. Much of our understanding about baleen whale population recovery and ecology depends on these methods 6-8. In oceanic regions close to population centres, these methods are often used to monitor regional population densities 8-10. However, regular applications of these methods are often constrained in remote, inaccessible regions, where their use represents a significant logistical and financial commitment 11. Consequently, such surveys are infrequent, making monitoring of population trends more challenging. In the Southern Ocean, the only comprehensive surveys south of 60° S (i.e. the putative summer foraging area for a range of cetacean species) were those undertaken by the International Whaling Commission (IWC) during the International Decade of Cetacean Research and the Southern Ocean Whale Ecosystem Research (IDCR SOWER) surveys, between 1978/9 and 2003/4. These surveys circumnavigated the continent three times, and based on these data Southern Ocean baleen whale recovery trends have been estimated 6,12,13. However, smallscale, sometimes ad hoc studies are far more common. These are generally biased towards the most accessible regions of the Southern Ocean 14 , the Western Antarctic Peninsula 3,4,15-18 , with more limited studies also conducted in the Scotia Arc 19 , Weddell Sea 20 and limited areas of East Antarctica 21,22. The Southern Ocean represents

show abstract

“…Researchers have explored both manual [ 10 , 11 , 12 , 27 ] and automated [ 28 , 29 ] methods of analysis ( Figure 1 ). Manually counting whales in satellite images is currently the most accurate method, although the most time-consuming, as it can require approximately 3 h and 20 min to scan 100 km 2 [ 12 ], and be erroneous due to observer bias [ 30 ].…”

Section: Considerations and Challenges Inherent To Satellite Imagementioning

confidence: 99%

“…Given the current limits on the available data of labelled satellite images of whales and similarly labelled images of confounding features within satellite imagery (e.g., rocks, boats, and white caps), required to create such a training dataset, aerial imagery can be used as a substitute for satellite imagery, to get the process underway. These aerial images should be down-sampled to the spatial resolution of VHR satellite imagery [ 28 ]. Alternatively, freely available photo archives of aerial and satellite imagery, can be accessed from sources such as Google Earth [ 29 ].…”

Section: Considerations and Challenges Inherent To Satellite Imagementioning

confidence: 99%

The Potential of Satellite Imagery for Surveying Whales

Höschle

Cubaynes

Clarke

et al. 2021

Sensors

Self Cite

View full text Add to dashboard Cite

The emergence of very high-resolution (VHR) satellite imagery (less than 1 m spatial resolution) is creating new opportunities within the fields of ecology and conservation biology. The advancement of sub-meter resolution imagery has provided greater confidence in the detection and identification of features on the ground, broadening the realm of possible research questions. To date, VHR imagery studies have largely focused on terrestrial environments; however, there has been incremental progress in the last two decades for using this technology to detect cetaceans. With advances in computational power and sensor resolution, the feasibility of broad-scale VHR ocean surveys using VHR satellite imagery with automated detection and classification processes has increased. Initial attempts at automated surveys are showing promising results, but further development is necessary to ensure reliability. Here we discuss the future directions in which VHR satellite imagery might be used to address urgent questions in whale conservation. We highlight the current challenges to automated detection and to extending the use of this technology to all oceans and various whale species. To achieve basin-scale marine surveys, currently not feasible with any traditional surveying methods (including boat-based and aerial surveys), future research requires a collaborative effort between biology, computation science, and engineering to overcome the present challenges to this platform’s use.

show abstract

Aerial-trained deep learning networks for surveying cetaceans from satellite imagery

Cited by 52 publications

References 44 publications

Using Deep Learning to Count Albatrosses from Space: Assessing Results in Light of Ground Truth Uncertainty

Using Deep Learning to Count Albatrosses from Space: Assessing Results in Light of Ground Truth Uncertainty

A comparison of baleen whale density estimates derived from overlapping satellite imagery and a shipborne survey

The Potential of Satellite Imagery for Surveying Whales

Contact Info

Product

Resources

About