Robotic Coral Reef Health Assessment Using Automated Image Analysis

Manderson, Travis; Li, Jimmy; Dudek, Natasha; Meger, David; Dudek, Gregory

doi:10.1002/rob.21698

Cited by 20 publications

(15 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More recent approaches are shifting to semantic segmentation, which is able to give more detailed information (pixel‐level) than the only classification. The first approaches performed image patch classification to thereafter reconstruct the segmentation of the entire image (Manderson et al, ; Shihavuddin et al, ). These kinds of patch‐based approaches, however, typically have low accuracy near the edges of the segmented regions.…”

Section: Related Workmentioning

confidence: 99%

“…When a few annotated pixels are provided, a CNN can be trained on patches cropped around those labeled pixels to get a final image segmentation joining the classification result for each patch. This strategy, which has been successfully applied in existing approaches (Beijbom et al, ; Manderson et al, ), is trained on

n

‐labeled patches, one per annotation. The training pairs used are of the form

(X_{d (i, j)}, y_{(i, j)})

where

X_{d (i, j)}

is a patch of dimensions

d \times d

centered around each labeled pixel with coordinates

(i, j)

, and

y_{(i, j)}

is a scalar representing the label of this pixel.…”

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

“…Regarding the Eilat data set, we compare our approach with prior work published by the authors of the data set for multiclass semantic segmentation. The authors (Beijbom et al, ) perform a patch‐based classification approach (explained in Section , the same approach that other works have followed; Manderson et al, ). We also compare our approach to our baseline and previous work (Alonso et al, ).…”

Section: Evaluation Of Our Proposed Approachmentioning

confidence: 99%

“…In particular, this study cropped image patches around the sparse labels and then performed image classification using support vector machine methods. Other works performed coral reef analysis using machine learning methods such as k-nearest neighbors (Manderson, Li, Dudek, Meger, & Dudek, 2017;Mary & Dharma, 2017;Shihavuddin, Gracias, Garcia, Gleason, & Gintert, 2013). Nevertheless, as previously mentioned, deep learning approaches are achieving state-of-the-art performance in classification, detection, and segmentation tasks (Garcia-Garcia et al, 2017), including coral reefs analysis (Moniruzzaman, Islam, Bennamoun, & Lavery, 2017).…”

Section: Coral Reef Community Structure Analysismentioning

confidence: 99%

“…As previously mentioned, many different projects ranging from autonomous surveys of coral reef ecosystems (Beijbom et al, 2012;Manderson et al, 2017) to wildlife monitoring from aerial systems (Hodgson, Baylis, Mott, Herrod, & Clarke, 2016) focus on monitoring tasks and subsequent data analysis. To enable automatic processing of the data, semantic segmentation models for the different target domains are needed, but their use is often blocked or hampered due to the lack of dense labeling to train semantic segmentation models, especially in domains where an expert is needed to label the images.…”

Section: Lack Of Training Datamentioning

confidence: 99%

See 4 more Smart Citations

CoralSeg: Learning coral segmentation from sparse annotations

Alonso

Yuval

Eyal

et al. 2019

Journal of Field Robotics

View full text Add to dashboard Cite

Robotic advances and developments in sensors and acquisition systems facilitate the collection of survey data in remote and challenging scenarios. Semantic segmentation, which attempts to provide per‐pixel semantic labels, is an essential task when processing such data. Recent advances in deep learning approaches have boosted this task's performance. Unfortunately, these methods need large amounts of labeled data, which is usually a challenge in many domains. In many environmental monitoring instances, such as the coral reef example studied here, data labeling demands expert knowledge and is costly. Therefore, many data sets often present scarce and sparse image annotations or remain untouched in image libraries. This study proposes and validates an effective approach for learning semantic segmentation models from sparsely labeled data. Based on augmenting sparse annotations with the proposed adaptive superpixel segmentation propagation, we obtain similar results as if training with dense annotations, significantly reducing the labeling effort. We perform an in‐depth analysis of our labeling augmentation method as well as of different neural network architectures and loss functions for semantic segmentation. We demonstrate the effectiveness of our approach on publicly available data sets of different real domains, with the emphasis on underwater scenarios—specifically, coral reef semantic segmentation. We release new labeled data as well as an encoder trained on half a million coral reef images, which is shown to facilitate the generalization to new coral scenarios.

show abstract

Section: Related Workmentioning

confidence: 99%