Optical coherence tomography (OCT) is a prevalent, interferometric, high-resolution imaging method with broad biomedical applications. Nonetheless, OCT images suffer from an artifact called speckle, which degrades the image quality. Digital filters offer an opportunity for image improvement in clinical OCT devices, where hardware modification to enhance images is expensive. To reduce speckle, a wide variety of digital filters have been proposed; selecting the most appropriate filter for an OCT image/image set is a challenging decision, especially in dermatology applications of OCT where a different variety of tissues are imaged. To tackle this challenge, we propose an expandable learnable despeckling framework, we call LDF. LDF decides which speckle reduction algorithm is most effective on a given image by learning a figure of merit (FOM) as a single quantitative image assessment measure. LDF is learnable, which means when implemented on an OCT machine, each given image/image set is retrained and its performance is improved. Also, LDF is expandable, meaning that any despeckling algorithm can easily be added to it. The architecture of LDF includes two main parts: (i) an autoencoder neural network and (ii) filter classifier. The autoencoder learns the FOM based on several quality assessment measures obtained from the OCT image including signal-to-noise ratio, contrast-to-noise ratio, equivalent number of looks, edge preservation index, and mean structural similarity index. Subsequently, the filter classifier identifies the most efficient filter from the following categories: (a) sliding window filters including median, mean, and symmetric nearest neighborhood, (b) adaptive statistical-based filters including Wiener, homomorphic Lee, and Kuwahara, and (c) edge preserved patch or pixel correlation-based filters including nonlocal mean, total variation, and block matching three-dimensional filtering.

Three approaches for instantaneous wide-area analysis of ship-radiated underwater sound, each focusing on a different aspect of that sound, received on a large-aperture densely-sampled coherent hydrophone array have been developed. (i) Ship’s narrowband machinery tonal sound is analyzed via temporal coherence using Mean Magnitude-Squared Coherence (MMSC) calculations. (ii) Ship’s broadband amplitude-modulated cavitation noise is examined using Cyclic Spectral Coherence (CSC) analysis that provides estimates for propeller blade pass rotation frequency, shaft rotation frequency, and hence the number of propeller blades. (iii) Mean power spectral densities (PSD) averaged across broad bandwidths are calculated in order to detect acoustically energetic ships. Each of these techniques are applied after beamforming of the received acoustic signals on a coherent hydrophone array, leading to significantly enhanced signal-to-noise ratios for simultaneous detection, bearing-time estimation and acoustic signature characterization of multiple ships over continental-shelf scale regions. The approaches are illustrated with underwater recordings of a 160-element coherent hydrophone array for six ocean vessels, that are located at a variety of bearings and ranges out to 200 km from the array, in the Norwegian Sea in February 2014. The CSC approach is shown to also be useful for automatic detection and bearing-time estimation of repetitive marine mammal vocalizations, providing estimates for inter-pulse-train and inter-pulse intervals from CSC spectra cyclic fundamental and first recurring peak frequencies respectively.

Humpback whale behavior, population distribution and structure can be inferred from long term underwater passive acoustic monitoring of their vocalizations. Here we develop automatic approaches for classifying humpback whale vocalizations into the two categories of song and non-song, employing machine learning techniques. The vocalization behavior of humpback whales was monitored over instantaneous vast areas of the Gulf of Maine using a large aperture coherent hydrophone array system via the passive ocean acoustic waveguide remote sensing technique over multiple diel cycles in Fall 2006. We use wavelet signal denoising and coherent array processing to enhance the signal-to-noise ratio. To build features vector for every time sequence of the beamformed signals, we employ Bag of Words approach to time-frequency features. Finally, we apply Support Vector Machine (SVM), Neural Networks, and Naive Bayes to classify the acoustic data and compare their performances. Best results are obtained using Mel Frequency Cepstrum Coefficient (MFCC) features and SVM which leads to 94% accuracy and 72.73% F1-score for humpback whale song versus non-song vocalization classification, showing effectiveness of the proposed approach for real-time classification at sea.

Humpback whale breathing-related sounds were recorded on elements of a coherent hydrophone array subaperture deployed vertically at the Great South Channel on the US Northeastern continental shelf in Fall 2021, where half of the hydrophones were in-air and the rest submerged underwater. In-air hydrophones recorded breathing sounds with approximately 2.5 s duration, but smaller bandwidths compared to underwater hydrophones where signal energies extended beyond 50 kHz, and a mean underwater source level of 161 ± 4 dB re 1 μPa at 1 m, based on measurements at 22.9 m. The underwater recorded humpback whale breathing sound spectra displayed a broadband dip centered at 15.7 kHz, with approximately 400 Hz half-power bandwidth, likely caused by attenuation from propagation through pulsating air bubbles. The air bubble radius for natural frequency of oscillations at 15.7 kHz is estimated to be 0.205–0.21 mm. These bubbles are capable of removing energy from the forward propagated humpback breathing sounds via resonance absorption most pronounced at and near bubble natural oscillation frequency. Humpback whale distances from the vertically deployed hydrophones are estimated and tracked by matching the curved nonlinear travel-time wavefront of its breathing sounds, since the whale was in the near-field of the subarray.

Developing automatic algorithms for real-time monitoring of underwater acoustic events is essential in ocean acoustic applications. Most previous ocean acoustic ecosystem monitoring studies are non-real-time, focus on data received on a single hydrophone or a specific analysis, such as bearing estimation or detection, without considering the full end-to-end analysis system. Here, we develop a unified framework for real-time ocean acoustic data analysis including beamforming, detection, bearing estimation and classification of transient underwater acoustic events. To detect sound sources, thresholding on computed mel-scale per-channel energy normalization is applied, followed by morphological image opening to extract pixels with significant intensities. Next, connected component analysis is applied for grouping pixel detections. The bearing of signal detections are next estimated via non-maximum suppression of 3D stacked beamformed spectrogram imageries. To classify a variety of whale species from their calls, time-frequency features are extracted from each detected signal’s beamformed power spectrogram. These features are next applied to train three classifiers, including Support Vector Machine, Neural Networks, and Random Forest to classify six whale vocalization categories: Fin, Sei, Unidentified Baleen, Minke, Humpback, and general Odontocetes. Best results are obtained with the Random Forest classifier, which achieved 96.7% accuracy, and 87.5% F1 score. A variety of accelerating approaches and fast algorithms are implemented to run on GPU. During an experiment in the U.S. Northeast coast in September 2021, the software and hardware advances developed here were utilized for near real-time analysis of underwater acoustic data received by Northeastern University’s in-house fabricated 160-element coherent hydrophone array system.

Image segmentation with clustering approach is widely used in biomedical application. Fuzzy c-means (FCM) clustering is able to preserve the information between tissues in image, but not taking spatial information into account, makes segmentation results of the standard FCM sensitive to noise. To overcome the above shortcoming, a modified FCM algorithm for MRI brain image segmentation is presented in this paper. The algorithm is realized by incorporating the spatial neighborhood information into the standard FCM algorithm and modifying the membership weighting of each cluster by smoothing it by Total Variation (TV) denoising. The proposed algorithm is evaluated with accuracy index in performing it on artificial synthesized images, and the results show the superior accuracy compared to some other state of the art FCM-based segmentation methods.