Hugo Robotham

Support vector machine under uncertainty: An application for hydroacoustic classification of fish-schools in Chile

In this work we apply multi-class support vector machines (SVMs) and a multi-class stochastic SVM formulation to the classification of fish schools of three species: anchovy, common sardine, and Jack Mackerel, and we compare their performance. The data used come from acoustic measurements in southern-central Chile. These classifications were carried out by using a diver set of descriptors including morphology, bathymetry, energy, and space positions. In both type of formulations, the deterministic and the stochastic one, the strategy used to classify multi-class SVM consists in employing the criterion one-species-against-the-Rest. We thus provide an empirical way to adjust the parameters involved in the stochastic classifiers with the aim of improving its performance. When this procedure is applied to the classification of fish schools we obtain a classifier with a better performance than the deterministic classifier.

Clasificación acústica de anchoveta (Engraulis ringens) y sardina común (Strangomera bentincki) mediante máquinas de vectores soporte en la zona centro-sur de Chile: efecto de la calibración de los parámetros en la matriz de confusión

RESUMEN. Se clasifico la anchoveta (Engraulis ringens) y sardina comun (Strangomera bentincki) detectadas mediante equipos acusticos en la zona centro-sur de Chile, mediante el metodo de Maquinas de Vectores Soporte (SVM). Para esto se utilizaron descriptores de cardumenes extraidos desde ecogramas, que fueron clasificados como morfologicos, batimetricos, energeticos y posicional espacial. Para lograr clasificaciones precisas mediante la utilizacion de esta metodologia, fue necesario optimizar parametros correspondientes al Kernel-Gaussiano, γ y de penalizacion del modelo C, mediante el analisis del efecto de la calibracion sobre las matrices de confusion resultantes de la clasificacion de las especies analizadas. El metodo SVM ajusto correctamente el 95,3% de los cardumenes de anchoveta y sardina comun. Los parametros optimos del Kernel-Gaussiano γ y de penalizacion C obtenidos mediante la metodologia propuesta fueron γ = 450 y C = 0,95, respectivamente. Los parametros mencionados incidieron de manera importante en la matriz de confusion y los porcentajes de clasificacion final, por lo que se sugiere establecer, en aplicaciones futuras de este metodo, un protocolo experimental de calibracion. La sardina comun fue la especie con menor error de clasificacion en el conjunto de las matrices de confusion. El descriptor correspondiente a profundidad del fondo fue el mas sensible al SVM, la segunda variable en importancia es el descriptor distancia a la costa. Palabras clave: maquinas de vectores soporte, clasificacion de especies, hidroacustica, peces pelagicos, anchoveta, sardina, Chile. Acoustic classification of anchovy (Engraulis ringens) and sardine (Strangomera bentincki) using support vector machines in central-southern Chile: effect of parameter calibration on the confusion matrix

Application of classifiers: Support vector machines, artificial neural networks and classification trees to identify acoustic schools

The purpose of this study was to compare the results of the classification of the pelagic fish species, the common sardine, anchovy, and jack mackerel with classification trees (CART), Support Vector Machine (SVM) and artificial neural network (multilayer perceptron, MLP), using mono-frequency acoustic data in southern-central Chile. The classifiers had similar performances, those of the MLP and SVM being the same, while t hat of CART was the lowest. The separation of anchovy and common sardine is considered acceptable with all methods, 90.8% for anchovy and between 87.4% (CART) and 90.3% (MLP) for sardine. These performances were higher than that for the jack mackerel, 77.8% (CART), 81.5% (MLP) and 85.2% (SVM). There is concordance on the groups of descriptors (bathymetric and positional) considered as effective for classification in all methods, but the importance of the descriptors presented by each method is not fully concordant. The energetic and morphological descriptor had low incidence. We recommend trying many classifiers to identify acoustic schools as a good practice.