Comparison of multifrequency acoustic and in situ measurements of zooplankton abundances in Knight Inlet, British Columbia

Enhanced convolutional neural network for plankton identification and enumeration

Despite the rapid increase in the number and applications of plankton imaging systems in marine science, processing large numbers of images remains a major challenge due to large variations in image content and quality in different marine environments. We constructed an automatic plankton image recognition and enumeration system using an enhanced Convolutional Neural Network (CNN) and examined the performance of different network structures on automatic plankton image classification. The procedure started with an adaptive thresholding approach to extract Region of Interest (ROIs) from in situ plankton images, followed by a procedure to suppress the background noise and enhance target features for each extracted ROI. The enhanced ROIs were classified into seven categories by a pre-trained classifier which was a combination of a CNN and a Support Vector Machine (SVM). The CNN was selected to improve feature description and the SVM was utilized to improve classification accuracy. A series of comparison experiments were then conducted to test the effectiveness of the pre-trained classifier including the combination of CNN and SVM versus CNN alone, and the performance of different CNN models. Compared to CNN model alone, the combination of CNN and SVM increased classification accuracy and recall rate by 7.13% and 6.41%, respectively. Among the selected CNN models, the ResNet50 performed the best with accuracy and recall at 94.52% and 94.13% respectively. The present study demonstrates that deep learning technique can improve plankton image recognition and that the results can provide useful information on the selection of different CNN models for plankton recognition. The proposed algorithm could be generally applied to images acquired from different imaging systems.

A semi-automated image analysis procedure for in situ plankton imaging systems

Plankton imaging systems are capable of providing fine-scale observations that enhance our understanding of key physical and biological processes. However, processing the large volumes of data collected by imaging systems remains a major obstacle for their employment, and existing approaches are designed either for images acquired under laboratory controlled conditions or within clear waters. In the present study, we developed a semi-automated approach to analyze plankton taxa from images acquired by the ZOOplankton VISualization (ZOOVIS) system within turbid estuarine waters, in Chesapeake Bay. When compared to images under laboratory controlled conditions or clear waters, images from highly turbid waters are often of relatively low quality and more variable, due to the large amount of objects and nonlinear illumination within each image. We first customized a segmentation procedure to locate objects within each image and extracted them for classification. A maximally stable extremal regions algorithm was applied to segment large gelatinous zooplankton and an adaptive threshold approach was developed to segment small organisms, such as copepods. Unlike the existing approaches for images acquired from laboratory, controlled conditions or clear waters, the target objects are often the majority class, and the classification can be treated as a multi-class classification problem. We customized a two-level hierarchical classification procedure using support vector machines to classify the target objects (< 5%), and remove the non-target objects (> 95%). First, histograms of oriented gradients feature descriptors were constructed for the segmented objects. In the first step all non-target and target objects were classified into different groups: arrow-like, copepod-like, and gelatinous zooplankton. Each object was passed to a group-specific classifier to remove most non-target objects. After the object was classified, an expert or non-expert then manually removed the non-target objects that could not be removed by the procedure. The procedure was tested on 89,419 images collected in Chesapeake Bay, and results were consistent with visual counts with >80% accuracy for all three groups.

Assessing the low frequency acoustic characteristics of Macrocystis pyrifera, Egregia menziessi, and Laminaria solidungula

The acoustic properties of kelp forests are not well known, but are of interest for the development of environmental remote sensing applications. This study examined the low-frequency (0.2-4.5 kHz) acoustic properties of three species of kelp (Macrocystis pyrifera, Egregia menziessi, and Laminaria solidungula) using a one-dimensional acoustic resonator. Acoustic observations and measurements of kelp morphology were then used to test the validity of Wood's multi-phase medium model in describing the acoustic behavior of the kelp. For Macrocystis and Egregia, the two species of kelp possessing pneumatocysts, the change in sound speed was highly dependent on the volume of free air contained in the kelp. The volume of air alone, however, was unable to predict the effective sound speed of the multi-phase medium using a simple two-phase (air + water) form of Wood's model. A separate implementation of this model (frond + water) successfully yielded the acoustic compressibility of the frond structure for each species (Macrocystis = 1.39 ± 0.82 × 10(-8) Pa(-1); Egregia = 2.59 ± 5.75 × 10(-9) Pa(-1); Laminaria = 8.65 ± 8.22 × 10(-9) Pa(-1)). This investigation demonstrates that the acoustic characteristics of kelp are species-specific, biomass-dependent, and differ between species with and without pneumatocyst structures.

Anisotropy in high-frequency broadband acoustic backscattering in the presence of turbulent microstructure and zooplankton

High-frequency broadband (120-600 kHz) acoustic backscattering measurements have been made in the vicinity of energetic internal waves. The transducers on the backscattering system could be adjusted so as to insonify the water-column either vertically or horizontally. The broadband capabilities of the system allowed spectral classification of the backscattering. The distribution of spectral shapes is significantly different for scattering measurements made with the transducers oriented horizontally versus vertically, indicating that scattering anisotropy is present. However, the scattering anisotropy could not be unequivocally explained by either turbulent microstructure or zooplankton, the two primary sources of scattering expected in internal waves. Daytime net samples indicate a predominance of short-aspect-ratio zooplankton. Using zooplankton acoustic scattering models, a preferential orientation of the observed zooplankton cannot explain the measured anisotropy. Yet model predictions of scattering from anisotropic turbulent microstructure, with inputs from coincident microstructure measurements, were not consistent with the observations. Possible explanations include bandwidth limitations that result in many spectra that cannot be unambiguously attributed to turbulence or zooplankton based on spectral shape. Extending the acoustic bandwidth to cover the range from 50 kHz to 2 MHz could help improve identification of the dominant sources of backscattering anisotropy.

Multifrequency species classification of acoustic-trawl survey data using semi-supervised learning with class discovery

Acoustic surveys often use multifrequency backscatter to estimate fish and plankton abundance. Direct samples are used to validate species classification of acoustic backscatter, but samples may be sparse or unavailable. A generalized Gaussian mixture model was developed to classify multifrequency acoustic backscatter when not all species classes are known. The classification, based on semi-supervised learning with class discovery, was applied to data collected in the eastern Bering Sea during summers 2004, 2007, and 2008. Walleye pollock, euphausiids, and two other major classes occurring in the upper water column were identified.

Response of Euphausia pacifica to small-scale shear in turbulent flow over a sill in a fjord

Zooplankton in the ocean respond to visual and hydro-mechanical cues such as small-scale shear in turbulent flow. In addition, they form strong aggregations where currents intersect sloping bottoms. Strong and predictable tidal currents over a sill in Knight Inlet, Canada, make it an ideal location to investigate biological behaviour in turbulent cross-isobath flow. We examine acoustic data (38, 120 and 200 kHz) collected there during the daylight hours, when the dominant zooplankters, Euphausia pacifica have descended into low light levels at ∼90 m. As expected, these data reveal strong aggregations at the sill. However, they occur consistently 10-20 m below the preferred light depth of the animals. We have constructed a simple model of the flow to investigate this phenomenon. Tracks of individual animals are traced in the flow and a variety of zooplankton behaviours tested. Our results indicate that the euphausiids must actively swim downward when they encounter the bottom boundary layer (bbl) to reproduce the observed downward shift in aggregation patterns. We suggest that this behaviour is cued by the small-scale shear in the bbl. Furthermore, this behaviour is likely to enhance aggregations found in strong flows at sills and on continental shelves.

Interpretation of multi-frequency acoustic data: effects of fish orientation

One goal of fisheries acoustics is to develop objective classification or identification methods to automate allocation of acoustic backscatter to species. Classification schemes rely on consistent relationships for successful apportionment of acoustic backscatter to species. A method is developed that compares frequency-dependent volume backscatter from an acoustical survey of Atlantic herring (Clupea harengus) to investigate the potential for classifying herring. Predicted backscattering patterns by a Kirchhoff-ray approximation are used to explain the observed relationships and evaluate the potential for classification of multi-frequency data. Combining predicted backscatter with observations of the frequency-dependent volume backscatter gave approximately 40% classification success, which is not sufficient for survey purposes. However, this method highlighted potential consequences that fish orientation may have on classification schemes and density and abundance estimates. This method of comparing multi-frequency volume backscatter appears to be beneficial for detecting behavioral changes by groups of fish, which may be used to select target strength values for density or abundance estimates. Utilizing predicted target strengths from numerical or analytical solutions or approximations, appropriate target strengths could be selected and would provide more accurate estimates of fish density and abundance.

Determining dominant scatterers of sound in mixed zooplankton populations

High-frequency acoustic scattering techniques have been used to investigate dominant scatterers in mixed zooplankton populations. Volume backscattering was measured in the Gulf of Maine at 43, 120, 200, and 420 kHz. Zooplankton composition and size were determined using net and video sampling techniques, and water properties were determined using conductivity, temperature, and depth sensors. Dominant scatterers have been identified using recently developed scattering models for zooplankton and microstructure. Microstructure generally did not contribute to the scattering. At certain locations, gas-bearing zooplankton, that account for a small fraction of the total abundance and biomass, dominated the scattering at all frequencies. At these locations, acoustically inferred size agreed well with size determined from the net samples. Significant differences between the acoustic, net, and video estimates of abundance for these zooplankton are most likely due to limitations of the net and video techniques. No other type of biological scatterer ever dominated the scattering at all frequencies. Copepods, fluid-like zooplankton that account for most of the abundance and biomass, dominated at select locations only at the highest frequencies. At these locations, acoustically inferred abundance agreed well with net and video estimates. A general approach for the difficult problem of interpreting high-frequency acoustic scattering in mixed zooplankton populations is described.