Classification of fish species from different ecosystems using the near infrared diffuse reflectance spectra of otoliths

Abstract

<jats:p> Applications of Fourier transform near infrared (FT-NIR) spectroscopy in fisheries science are currently limited. This current analysis of otolith spectral data demonstrate the potential applicability of FT-NIR spectroscopy to otolith chemistry and spatial variability in fisheries science. The objective of this study was to examine the use of NIR spectroscopy as a tool to differentiate among marine fishes in four large marine ecosystems. We examined otoliths from 13 different species, with three of these species coming from different regions. Principal component analysis described the main directions along which the specimens were separated. The separation of species and their ecosystems may suggest interactions between fish phylogeny, ontogeny, and environmental conditions that can be evaluated using NIR spectroscopy. In order to discriminate spectra across ecosystems and species, four supervised classification model techniques were utilized: soft independent modelling of class analogies, support vector machine discriminant analysis, partial least squares discriminant analysis, and k-nearest neighbor analysis (KNN). This study showed that the best performing model to classify combined ecosystems, all four ecosystems, and species was the KNN model, which had an overall accuracy rate of 99.9%, 97.6%, and 91.5%, respectively. Results from this study suggest that further investigations are needed to determine applications of NIR spectroscopy to otolith chemistry and spatial variability. </jats:p>

Assessing the Ecological and Biological Forage Base of the Tobol River: Implications for Planktonic and Benthic Communities

“…During the period from 2016 to 2019, 36 species and forms of benthic organisms were registered in the Tobol River basin. There is a great variety of insect groups such as hemipterans, dragonflies and coleopterans [20]. The high diversity of organisms is primarily due to the considerable length of the river, located in various landscape zones and environmental conditions.…”

Enhancing predictive performance for spectroscopic studies in wildlife science through a multi-model approach: A case study for species classification of live amphibians

“…A benchmark modeling exercise was conducted in the open-source software, R, in a multi-class problem to compare the performance of seven conventional machine learning algorithms: generalized linear model with elastic net regression (GLM), K-nearest neighbors (KNN), linear discriminant analysis (LDA), partial least squares (PLS), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB). The scope of this study is limited to the aforementioned algorithms, but many others have recently been applied (e.g., Naïve Bayes, neural networks) to generate prediction models for spectroscopic datasets [79][80][81]. Each model technique adheres to a unique, established set of mathematical rules to first assess spectral variability present in the dataset and then calibrate a model to subsequently make predictions on unseen test data.…”

“…Results presented in Table 1 include studies containing vertebrate species exclusively, and studies found in our literature search were excluded if they did not clearly specify the algorithm or model type used for generating prediction models. Additionally, our search was limited to studies utilizing shallow machine learning algorithms, although deep learning is an emerging predictive modeling technique in spectroscopic studies [80,[113][114][115].…”

“…PLS models have been applied to answer a diverse set of questions across many different sampling types, such as species discrimination from dried blood traces between Bengal tigers and Indian leopards, and quantification of dietary components from the feces of black bears [ 19 , 20 ]. However, it may be that different sample types and associated questions can be better addressed through the exploration of other algorithms [ 21 ]. The wide variety of sampling modes (e.g., feces, hair, otolith, live animals) and questions addressed in wildlife sciences warrants a multi-faceted approach to robustly generate prediction models that are best suited for the sample type under investigation.…”

Bayesian Fusion Model Enhanced Codfish Classification Using Near Infrared and Raman Spectrum

“…However, these methods are clearly time-consuming, destructive, unable to achieve rapid detection on site or require trained personnel [ 12 ]. In parallel, several spectroscopic techniques combined with chemometrics have been employed [ 13 , 14 ]. These studies have demonstrated the potential of vibration spectroscopy for rapid and non-destructive identity assessment on seafood.…”