An Investigation of Preprocessing Filters and Deep Learning Methods for Vessel Type Classification With Underwater Acoustic Data

May combine two claims.

Using all three preprocessing representations simultaneously (CQT, Gammatone, and Mel spectrogram) yielded the most reliable result, with over 97% accuracy.

“With over 97% accuracy, using all three preprocessing representations simultaneously yielded the most reliable result.”

May combine two claims.

The classification accuracy obtained using the three-channel Complete representation was marginally better than the best single filter (CQT) for ResNet, but not the VGG model.

“The results obtained showed that the classification accuracy obtained using this three-channel representation was marginally better than the best single filter (CQT) for ResNet, but not the VGG model.”

May combine two claims.

A minor degradation in accuracy was observed with respect to an increase in the distance to the sensor, but the results obtained for the three scenarios showed similar figures.

“Despite the fact that a minor degradation in accuracy was observed with respect to an increase in the distance to the sensor, the results obtained for the three scenarios showed similar figures.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Combining CQT, Gammatone, and Mel spectra resulted in data samples with dimensions of 64x63x3 for Version 1 and 95x128x3 for Version 2.

“Combining CQT, Gammatone, and Mel spectra resulted in data samples with dimensions of 64x63x3 for Version 1 and 95x128x3 for Version 2.”

May combine two claims.

The ResNet18 model outperformed VGG for both CQT and Complete preprocessing filters, presenting an improvement of 8.09 percentage points for the CQT, and 14.15 for the Complete.

“The results showed that the ResNet18 model outperformed VGG for both CQT and Complete preprocessing filters, presenting an improvement of 8.09 percentage points for the CQT, and 14.15 for the Complete, the latter being the best result obtained for this dataset scenario overall.”

May combine two claims.

SGD outperformed Adam for CQT, Mel, and the Complete representation, with Complete having the highest values.

“The results represented in Table 5 show that SGD outperformed Adam for CQT, Mel, and the Complete representation, where the latter had the highest values (as shown in bold font in Table 5).”

May combine two claims.

SGD produced higher accuracy and a more stable performance.

“In comparison, SGD produced higher accuracy and a more stable performance.”

May combine two claims.

The training session using Adam and the Mel spectrogram did not converge to a global minimum, as the model predicted that almost everything belonged to the same class.

“In addition, the training session using Adam and the Mel spectrogram did not converge to a global minimum as the model predicted that almost everything belonged to the same class, as per the class-normalised confusion matrix shown in Figure 6.”

May combine two claims.

Despite the accuracy drop of 1.54 percentage points between Version 1 and Version 2 for the Complete scenario, the precision improved 1.58 percentage points.

“Additionally, despite the accuracy drop of 1.54 percentage points between Version 1 and Version 2 for the Complete scenario, the precision improved 1.58 percentage points, showing a very close result for both versions of the Complete representation.”

May combine two claims.

The higher accuracy comparing best results for SGD and Adam with CQT and Complete inputs was obtained with the SGD optimiser (84.46%), which is 7 percentage points larger than the best result obtained with Adam (77.39%).

“Comparing the best results for SGD and Adam, with CQT and Complete inputs, the higher accuracy was obtained with the SGD optimiser (84.46%), which is 7 percentage points larger accuracy than the best result obtained with Adam (77.39%).”

May combine two claims.

The Complete representation combined with the ResNet model presented an improvement of 2.12 percentage points in accuracy over CQT, but CQT was able to obtain a similar value using only one-third of the input size and preprocessing.

“Although the Complete representation, combined with the ResNet model, presented an improvement of 2.12 percentage points in accuracy, the CQT was able to obtain a similar value using only one-third of the input size and preprocessing, since Complete is a three-channel representation.”

May combine two claims.

In our dataset, the CQT representation presented better performance, with an accuracy of around 86%, when compared with Gammatone and Mel spectrograms, with best accuracies of 62% and 54%, respectively, a minimum of 24 percentage points improvement.

“In our dataset, the CQT representation presented better performance, with an accuracy of around 86%, when compared with Gammatone and Mel spectrograms, with best accuracies of 62% and 54%, respectively, a minimum of 24 percentage points improvement.”

Classified as probably not a substantial claim.

Three distinct scenarios were defined with respect to the distance between the target vessel and the hydrophone used to capture the signal.

“Three distinct scenarios were defined with respect to the distance between the target vessel to the hydrophone used to capture the signal.”

• AssumptionRests on an assumption: The distances between target vessels and the hydrophone are accurately known for defining the three scenarios.

May combine two claims.

The three distance scenarios were compared using the proposed pipeline, achieving a maximum accuracy of 94.95% when CQT was used to preprocess the data fed into a ResNet classification model.

“These three scenarios were compared using the proposed pipeline, achieving a maximum accuracy of 94.95% when CQT was used to preprocess the data fed into a ResNet classification model.”

• AssumptionRests on an assumption: The vessel type annotations in the ONC dataset are correct.
• AssumptionRests on an assumption: The distances between target vessels and the hydrophone are accurately known for defining the three scenarios.

May combine two claims.

For distinct distance scenarios, the present work achieved an accuracy of 95% with CQT preprocessing or 97% with Complete (combined) preprocessing.

“However, the research reported here achieved superior results when distinct scenarios were considered with respect to the distance to the sensor, obtaining an accuracy of 95% or 97% depending on the preprocessing filtering method used (CQT or Complete, respectively), as shown in Table 9.”

• AssumptionRests on an assumption: The vessel type annotations in the ONC dataset are correct.
• AssumptionRests on an assumption: The distances between target vessels and the hydrophone are accurately known for defining the three scenarios.

Classified as probably not a substantial claim.

A complete pipeline for the classification of underwater acoustic signals was proposed in this paper.

“A complete pipeline for the classification of underwater acoustic signals was proposed in this paper, whose source code and data will be made publicly available 4 .”

May combine two claims.

The SGD optimiser is more stable and presents better generalisation than Adam for the vessel type classification problem.

“This study also compared the SGD and Adam optimiser performance applied to the vessel type classification problem, showing that the SGD optimiser is more stable and presents a better generalisation than Adam.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Combining multiple preprocessing filters as inputs to CNN models achieves superior outcomes compared to individual approaches.

“The comparison of the results obtained with individual approaches and their combination showed that superior outcomes could be achieved with the latter.”

• Also stated in: § II. RELATED WORK · plain

  “Another relevant work used a fusion of the Mel-spectrogram, MFCC, chromatogram, spectral contrast, and Tonnetz filters, resulting in a one-dimensional representation, to improve the performance of a CNN model for the classification of underwater acoustic signals [43].”

• cites [42]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [43]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

The results were further analysed and filtered to avoid mislabelling from the automatic annotation procedure.

“This automatic annotation procedure could generate mislabelling in the dataset, therefore the results were further analysed and filtered to avoid this issue.”

• AssumptionRests on an assumption: The automatic annotation procedure could generate mislabelling in the dataset.

Probably not a substantial claim: may not pass S3 (load-bearing).

Environmental conditions such as temperature, depth, salinity, pressure, and precipitation can directly influence how the signal travels from emitter to receiver.

“Environmental conditions, such as temperature, depth, salinity, pressure, and even precipitation, can directly influence how the signal travels from emitter to receiver [28].”

• cites [28]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Recent state-of-the-art methods for underwater acoustic signal analysis and classification are largely based on deep-learning algorithms.

“However, recent state-of-the-art methods are largely based on the application of deep-learning algorithms to solve similar tasks [17]- [19].”

• cites [17]- [19]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Underwater acoustic signals have a low signal-to-noise ratio (SNR) and a high degree of variability for the same source.

“it has a low signal-to-noise ratio (SNR) and a high degree of variability for the same source [15], [16]”

• cites [15]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [16]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• AssumptionRests on an assumption: Underwater acoustic signals are a mixture of environmental, biological, and human-generated sounds.

Probably not a substantial claim: may not pass S3 (load-bearing).

The task of underwater acoustic target classification is challenging due to the complex nature of the sound produced by vessels.

“The task of underwater acoustic target classification is challenging due to the complex nature of the sound produced by vessels [25].”

• cites [25]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• AssumptionRests on an assumption: Different types of vessels, in diverse states of upkeep, produce fundamentally distinct sounds depending on the vessel's speed, the state of its mechanical parts, and the hydrodynamics of its design.

Probably not a substantial claim: may not pass S3 (load-bearing).

Smaller CNN models can have a beneficial trade-off between processing time and accuracy.

“This proved that smaller models can have a beneficial trade-off between processing time and accuracy.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Classical signal processing methods such as Cepstral analysis can improve the quality of processed sound by reducing reflections interference and scattering losses, but only for short ranges with high SNR.

“some classical signal processing methods, such as Cepstral analysis [14], can improve the quality of the processed sound by reducing the effects of the reflections interference and scattering losses, but only if applied on signals for short ranges with a high SNR [29].”

• cites [14]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [29]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The acoustic signals emitted by vessels captured using hydrophones provide a low-cost and fraud-resistant data source for surveillance tasks, because acoustic signals can be difficult to omit or mask.

“the acoustic signals emitted by vessels captured using hydrophones provide a low-cost and fraud-resistant data source to be used in surveillance tasks, as acoustic signals can be difficult to omit or mask.”

May combine two claims.

Superior results, smaller networks, and reduced training time are obtained with the use of appropriate preprocessing filters in the classification pipeline.

“Therefore, superior results, in addition to smaller networks and reduced training time, are obtained with the use of appropriate preprocessing filters in the classification pipeline.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Additional complexity exists due to background sound produced by the region and the complexity of sound propagation in shallow waters, which causes multi-path reflections.

“additional complexity exists due to the background sound produced by the region and the complexity of sound propagation in shallow waters, which causes multi-path reflections [27].”

• cites [27]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Using preprocessing filters in underwater audio tasks yields better results than applying a CNN directly to raw audio or than baseline approaches.

“Similarly, Mel-log spectrograms, delta, and delta-delta features were also used as acoustic filters in a ship detection task using a CNN, where high accuracy in the detection and localisation of vessels was reported [37].”

• Also stated in: § II. RELATED WORK · boosted

  “Other studies in the literature also successfully applied filters to the DL inputs, showing a consistent improvement in underwater audio classification tasks [38]- [41].”

• cites [36]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [37]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [38]- [41]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The most recent studies have used VGGNet or ResNet models as base algorithms for this development, owing to the models' high accuracy in complex classification tasks.

“The most recent studies have used VGGNet [23] or ResNet [24] models as base algorithms for this development, owing to the models' high accuracy in complex classification tasks.”

• cites [23]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [24]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Although CNNs can be applied directly to the audio signal, acoustic filters are frequently used as preprocessing layers to improve the quality of the resulting audio representations.

“Although CNNs can be applied directly to the audio signal [31], [34], [35], acoustic filters are frequently used as preprocessing layers to improve the quality of the resulting audio representations [25].”

• cites [25]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [31]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [34]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

CNNs are now being considered for underwater acoustic classification applications.

“Advances in machine learning techniques mean CNNs are now being considered for underwater acoustic classification applications [25].”

• cites [25]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

Isolating data to scenarios where only a single vessel was within the inclusion zone and no other vessels were within the exclusion zone ensured a more reliable acoustic signature without interference from other vessels.

“Data was isolated where only a single vessel was within the inclusion zone while no other vessels were within the exclusion zone.This ensured a more reliable acoustic signature without interference from other vessels.”

Probably not a substantial claim: may not pass S2 (checkable).

This more accurate classifier for underwater acoustics could be used as a reliable autonomous monitoring system in maritime environments.

“This more accurate classifier for underwater acoustics could be used as a reliable autonomous monitoring system in maritime environments.”

• AssumptionRests on an assumption: There is a need for an autonomous monitoring system that can generate an alert whenever a particular class of vessel is detected in an area, in order to better monitor and protect marine environments.
• AssumptionRests on an assumption: Machine learning methods are useful in object detection and classification tasks and have the potential to underpin techniques for the development of robust monitoring systems to overcome the problem of illegal exploitation of protected marine environments.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

There is a pertinent need for a complete investigation into the application of DL algorithms for underwater acoustic classification that considers optimisation of DL model parameters and comparison between different preprocessing filters.

“However, there is a pertinent need for a complete investigation into the application of DL algorithms for the task, an investigation that considers the optimisation of the DL model parameters, and the comparison between different preprocessing filters.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Mel's scale is empirically based on the way humans perceive sound.

“Mel's scale is empirically based on the way humans perceive sound [47].”

• cites [47]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Because the sound generated by sources in the underwater domain is dependent on the environment, it is important that the type of two-dimensional representation matches the problem and that its parameters are optimised for the task.

“as the sound generated by sources in the underwater domain is dependent on the environment, it is of extreme importance that not only the type of two-dimensional representation matches the problem, but also its parameters must be optimised for the task.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Recent studies, mostly based on Deep Learning methods, have shown promise in automatic data classification tasks.

“recent studies, mostly based on the application of Deep Learning (DL) methods [17]- [19], have shown promise in automatic data classification tasks.”

• cites [17]- [19]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The engine typically produces a broadband continuous spectrum, while the propeller generates narrow band components with power at discrete frequencies.

“The former typically produces a broadband continuous spectrum, while the latter generates narrow band components whose spectrum consists of power at discrete frequencies [26].”

• cites [26]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

The rationale behind using stacks of multiple filters is to take advantage of the strengths of each method by feeding the network with different representations of the sound.

“The rationale behind this approach is to take advantage of the strengths of each method, feeding the network with different representations of the sound.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Deeper CNN models achieved less than 1% accuracy improvement over shallow models for mine-like object detection from sidescan sonar images, at the cost of a 17x increase in computational requirements.

“The comparative results reported suggest that deeper models (i.e., models with multiple layers) achieved less than 1% of accuracy improvement when compared with shallow models, at the cost of a 17x increase of computational requirements.”

• cites [30]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The gammatone filter was first defined as a filter bank capable of representing the shape of the impulse response function of the human auditory system.

“The gammatone filter was first defined as a filter bank capable of representing the shape of the impulse response function of the human auditory system [51].”

• cites [51]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

SGD is an iterative method that starts randomly and seeks the minimum value in the input function.

“SGD is an iterative method that starts randomly and seeks the minimum value in the input function.”

• cites [54]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The gammatone representation has an amplitude characteristic that can be used to predict the human auditory response.

“This representation has an amplitude characteristic that can be used to predict the human auditory response.”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

An accuracy value of around 87% was reported for a bio-inspired cochlea model preprocessing filter applied to a CNN-based classification.

“An accuracy value of around 87% was reported as a result of the application of a bio-inspired cochlea model preprocessing filter to a CNN-based classification [40].”

• cites [40]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The function defined in Equation (8) was used by [51] to summarise the RevCor, a representation of the correlation between a sound stimulus on the human ear and the response of a primary auditory fibre.

“The function defined on Equation ( 8) was used by [51] to summarise the RevCor, a representation of the correlation between a sound stimulus on the human ear and the response of a primary auditory fibre [52].”

• cites [51]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

The acoustic signals generated by the vessels are predominantly composed of lower frequencies.

“One is that the acoustic signals generated by the vessels are predominantly composed of lower frequencies.”

Classified as probably not a substantial claim. May combine two claims.

Although ResNet18 produced the best classification results, VGG-based classifiers are still suitable models to be used in applications with limited computational resources.

“Considering the relative complexity and accuracy results of both models tested in this work, we can conclude that, although ResNet18 produced the best classification results, VGG-based classifiers are still suitable models to be used in applications with limited computational resources.”

Classified as probably not a substantial claim. May combine two claims.

The depth and ocean temperature where the hydrophone was located provided the best context for underwater sound propagation, thereby not degrading the SNR sufficiently to invalidate the signal representation.

“One explanation for this could be the depth and ocean temperature where the hydrophone was located, which provided the best context for underwater sound propagation [25], thereby not degrading the SNR sufficiently to invalidate the signal representation.”

• cites [25]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

ResNet addresses the vanishing gradient and accuracy degradation issues by introducing the identity shortcut connection, bypassing one or more layers in a forward pass, defining Residual Blocks.

“ResNet addresses these issues by introducing the identity shortcut connection, bypassing one or more layers in a forward pass, defining Residual Blocks.”

• cites [24]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The different scenarios do not contain the same number of instances (or the same vessels), so they are not directly comparable.

“As the different scenarios do not contain the same number of instances (or the same vessels), they are not directly comparable, making it difficult to precisely state which situation allowed the best outputs.”

May combine two claims.

The worst results in the combined dataset were probably due to the fact that, although a minor variation in the SNR did not affect the results obtained in each of the individual scenarios, this difference was large enough to increase the complexity of the audio patterns contained in the combined dataset, thus hindering the capacity of a simple classifier to find a suitable generalisation that, in,

“This was probably due to the fact that, although a minor variation in the SNR (resulted from the distance between target and sensor) did not affect the results obtained in each of the individual scenarios, this difference was large enough to increase the complexity of the audio patterns contained in the combined dataset, thus hindering the capacity of a simple classifier to find a suitable general”

May combine two claims.

Prior work using ONC data reported accuracy values of the order of 80% using raw audio data, where time-frequency filter dependency was not considered in deep learning pipelines.

“Considering related work developed with data from the ONC initiative, accuracy values of the order of 80% were previously reported using raw audio data, where time-frequency filter dependency was not considered in DL pipelines [34], [35].”

• cites [34]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [35]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The gammatone representation has a minimum-phase characteristic, which is a preferred feature for an auditory filter bank.

“It also has a minimum-phase characteristic, which is a preferred feature for an auditory filter bank [52].”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Mel Spectrograms are commonly used in speech recognition analysis and music processing, where human perception is extremely relevant.

“Mel Spectrograms are commonly used in speech recognition analysis and music processing, where human perception is extremely relevant [49].”

• cites [49]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Correct tuning of the quality factor Q can supply the needed information for acoustic analysis, with resolution to distinguish adjacent musical notes, where a sound with harmonic frequency components will produce a constant pattern in the log frequency domain.

“The correct tuning of the quality factor Q can supply the needed information for the acoustic analysis, with resolution to distinguish adjacent musical notes, where a sound with harmonic frequency components will produce a constant pattern in the log frequency domain [50].”

• cites [50]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Adam is an extension of SGD based on the combination of the Adaptive Gradient Algorithm (AdaGrad) and the Root Mean Square Propagation (RMSP).

“Adam, on the other hand, is an extension of SGD based on the combination of the Adaptive Gradient Algorithm (AdaGrad) and the Root Mean Square Propagation (RMSP).”

• cites [55]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

CQT increases the time resolution towards higher frequencies while reducing the frequency resolution; this results in emphasising the lower-frequencies, which is akin to the human aural perception.

“In contrast, CQT increases the time resolution towards higher frequencies while reducing the frequency resolution; this results in emphasising the lower-frequencies, which is akin to the human aural perception [50].”

• cites [50]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The ResNet's capacity to have more intermediate layers proved to be suitable for the feature extraction stage, as it gave the model the ability to generalise the problem function better, thus resulting in higher classification performance.

“The ResNet's capacity to have more intermediate layers proved to be suitable for the feature extraction stage, as it gave the model the ability to generalise the problem function better, thus resulting in higher classification performance.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Residual blocks facilitate the learning ability of the intermediary layers, reducing the vanishing gradient problem, and penalising the ones that could potentially degrade accuracy.

“These blocks facilitate the learning ability of the intermediary layers, reducing the vanishing gradient problem, and penalising the ones that could potentially degrade accuracy [24].”

• cites [24]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The momentum-based strategy of SGD provides a drift effect to escape saddle points without affecting the flat minima selection.

“In contrast, the momentum-based strategy of the latter provides a drift effect to escape saddle points without affecting the flat minima selection [57].”

• cites [57]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The Gammatone filter does not emphasise low frequencies sufficiently, resulting in a lower classification performance.

“However, the Gammatone filter does not emphasise low frequencies sufficiently, resulting in a lower classification performance.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The work in [41] used a dataset called DeepShip, whose recordings were divided into 613 files varying from about 6 seconds to 1530 seconds.

“That work was developed on a dataset called DeepShip, whose recordings were divided into 613 files, which varies from about 6 seconds to 1530 seconds.”

• cites [41]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In the DeepShip dataset, only the identification of a single vessel within a range of 2 km from the hydrophone was used to generate the data, and background noise recordings were added from a distinct source.

“Only the identification of a single vessel within a range of 2 km from the hydrophone was used to generate the data, and the background noise recordings were added from a distinct source.”

• cites [41]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

A comparison of various deep learning methods was conducted in [41] using an analogous set of ONC raw data that was used in the present paper.

“In addition, a comparison of various deep learning methods was conducted in [41] using an analogous set of ONC raw data that was used in the present paper.”

• cites [41]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The highest accuracy obtained in [41] was 77.53%.

“However, it was reported that 77.53% was the highest accuracy obtained in that work.”

• cites [41]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.