Ismael F. Aymerich

A Rapid Technique for Classifying Phytoplankton Fluorescence Spectra Based on Self-Organizing Maps

Fluorescence spectroscopy has been demonstrated to be a powerful tool for characterizing phytoplankton communities in marine environments. Using different fluorescence spectra techniques, it is now possible to discriminate the major phytoplankton groups. However, most of the current techniques are based on fluorescence excitation measurements, which require stimulation at different wavelengths and thus considerable time to obtain the complete spectral profile. This requirement may be an important constraint for several mobile oceanographic platforms, such as vertical profilers or autonomous underwater vehicles, which require rapid-acquisition instruments. This paper presents a novel technique for classifying fluorescence spectra based on self-organizing maps (SOMs), one of the most popular artificial neural network (ANN) methods. The method is able to achieve phytoplankton discrimination using only fluorescence emission spectra (single wavelength excitation), thus reducing the acquisition time. The discrimination capabilities of SOM using excitation and emission spectra are compared. The analysis shows that the SOM has a good performance using excitation spectra, whereas data preprocessing is required in order to obtain similar discrimination capabilities using emission spectra. The final results obtained using emission spectra indicate that the discrimination is properly achieved even between algal groups, such as diatoms and dinoflagellates, which cannot be discriminated with previous methods. We finally point out that although techniques based on excitation spectra can achieve a better taxonomic accuracy, there are some applications that require faster acquisition processes. Acquiring emission spectra is almost instantaneous, and techniques such as SOM can achieve good classification performance using appropriately preprocessed data.

Effect of spectral resolution in hyperspectral data analysis

The effect of hyperspectral data resolution on the results obtained using derivative spectroscopy is discussed in this article. A comparison was made between attenuation spectra measured using two different hyperspectral sensors with 3648 and 256 spectral bands, respectively. Smoothing and derivative algorithms were applied to both types of spectral data in order to assess qualitative information from the spectral features of several laboratory samples. In order to make an optimal application of the derivative analysis, a suitable selection of the smoothing and derivating parameters was done according to the resolution of each type of hyperspectral data. Second derivative spectra were obtained and peaks related to the absorption bands of pigments present in the considered samples were identified in both cases. Furthermore, interpolation techniques were applied so as to match spectral resolutions of data collected by the two sensors and negligible variations in the positions of the peaks were achieved in the derivative spectra.

Monolithic spectrometer for environmental monitoring applications

Hyperspectral data analysis has been shown to be a reliable technique to identify several water components. Currently, there are numerous commercially available miniature spectrometer systems as well as discrete components that are used by researchers in designing their own systems. We have developed a custom-designed hyperspectral sensor. Experimental results and potential applications are discussed on this article. It can be used as a low cost optical monitoring system to be employed on fish farming and aquaculture facilities.

Hyperspectral remote sensing of phytoplankton assemblages in the ocean: Effects of the vertical distribution

The increasing availability of hyperspectral technology on remote sensing observing platforms provides marine scientists the potential to better map the phytoplankton community composition in the ocean globally. In this study, this approach is examined using unsupervised cluster techniques applied to data sets of hyperspectral remote sensing reflectance obtained by radiative simulations. Different oceanic environments in terms of phytoplankton biodiversity were satisfactorily classified. Furthermore, an assessment of the effect of a variable stratified water column scenario on the remote sensing reflectance points out that vertical distributions of phytoplankton communities along the water column play an essential role in the accurate phytoplankton assemblages mapping by remote sensing.

Analysis of Discrimination Techniques for Low-Cost Narrow-Band Spectrofluorometers

The need for covering large areas in oceanographic measurement campaigns and the general interest in reducing the observational costs open the necessity to develop new strategies towards this objective, fundamental to deal with current and future research projects. In this respect, the development of low-cost instruments becomes a key factor, but optimal signal-processing techniques must be used to balance their measurements with those obtained from accurate but expensive instruments. In this paper, a complete signal-processing chain to process the fluorescence spectra of marine organisms for taxonomic discrimination is proposed. It has been designed to deal with noisy, narrow-band and low-resolution data obtained from low-cost sensors or instruments and to optimize its computational cost, and it consists of four separated blocks that denoise, normalize, transform and classify the samples. For each block, several techniques are tested and compared to find the best combination that optimizes the classification of the samples. The signal processing has been focused on the Chlorophyll-a fluorescence peak, since it presents the highest emission levels and it can be measured with sensors presenting poor sensitivity and signal-to-noise ratios. The whole methodology has been successfully validated by means of the fluorescence spectra emitted by five different cultures.

Fast phytoplankton classification from fluorescence spectra: comparison between PSVM and SOM

Evaluation of phytoplankton communities is an important task to characterize marine environments. Fluorescence spectroscopy is a powerful technique usually used for this goal. This study presents a comparison between two different techniques for fast phytoplankton discrimination: Self-Organizing Maps (SOM) and Potential Support Vector Machines (P-SVM), evaluating its capability to achieve phytoplankton classification from its fluorescence spectra. Several cultures representing different algae groups were grown under the same conditions and their emission fluorescence spectra were measured every day. Finally, the classification results obtained from both techniques, SOM and P-SVM, are presented. In the case of using emission fluorescence spectra, the results show that we are able to reduce the acquisition time required for some of the existing techniques, obtaining encouraging classification performance.

Comparing the use of hyperspectral Irradiance Reflectance and Diffuse Attenuation Coeficient as indicators for algal presence in the water column

Remarkable advances are being achieved on spectroradiometric systems technology, allowing the development of spectrometers aimed at in-situ measurements. However, the use of numerical modelling of the ocean optical properties can help to understand how these properties react to changes in the water columns composition. In this work, HydroLight-EcoLight was used in order to simulate different scenarios to study how two Apparent Optical Properties (AOPs) such as Irradiance Reflectance (R) and Diffuse Attenuation Coefficient (Kd) could be used to detect the presence of a specific phytoplankton group in the water column.

Potential support vector machines and Self-Organizing Maps for phytoplankton discrimination

Fluorescence spectroscopy is a powerful technique usually used to evaluate phytoplankton marine environments. In this study, a kernel method (Potential Support Vector Machine, P-SVM) is presented, evaluating its capability to achieve phytoplankton classification from its fluorescence spectra. Different phytoplankton species were studied, and their fluorescence spectra were acquired in laboratory. In a previous study working with Self-Organizing Maps (SOM), it was proved with experimental data from laboratory that excitation spectra were more discriminative than emission spectra. It was also shown that using some preprocessing techniques, such as derivative analysis, the classification performance from emission fluorescence data can be improved. The classification results were encouraging to keep working with emission fluorescence, and herein we present a comparison between P-SVM and SOM for this goal.

Mapping marine phytoplankton assemblages from a hyperspectral and Artificial Intelligence perspective

The aim of this contribution is to demonstrate the feasibility of different processing techniques to identify phytoplankton assemblages when applied to oceanographic hyperspectral data sets (i.e. above surface measurements and vertical profiles). In order to address this issue and validate the proposed techniques, a simulated framework has been used based on the oceanic radiative transfer model Hydrolight-Ecolight 5.0. The potential offered by an unsupervised hierarchical cluster analysis technique and two Artificial Intelligence algorithms (i.e. Particle Swarm Optimization and Case-Based Reasoning) have been explored. Our results confirm their suitability to map phytoplanktons distribution from hyperspectral information given a variety of hypothetical oceanic environments.

Fast phytoplankton classification from emission fluorescence spectra based on Self-Organizing Maps

Fluorescence spectroscopy is a powerful technique usually used to evaluate phytoplankton marine environments. In this study, a fast-technique for phytoplankton discrimination is presented based on the Self-Organizing Maps (SOM), evaluating its capability to achieve phytoplankton classification from its emission fluorescence spectra. The aim of this work is to reduce the acquisition time required for some of the existing techniques. Several cultures representing different algae groups were grown under the same conditions and their Emission spectra were measured every day. Finally, SOM analysis combined with derivative analysis was performed obtaining encouraging results.