Linilsson Rodrigues Padovese

This article explores the application of machine learning techniques in acoustic ecology to classify the formations of the Brazilian Cerrado (Forest, Savanna, and Grassland) based on their soundscapes. Considering the Cerrado’s importance for biodiversity and hydrology, as well as the challenges faced by the biome in the face of agricultural expansion, the study seeks more efficient and economical methods for identifying its physiognomies. Five statistical models were developed and evaluated, using both traditional Machine Learning and Deep Learning, with the use of Mel-Frequency Cepstral Coefficients (MFCCs) and spectrogram images as input variables. The performance of these models was measured by accuracy, precision, and recall metrics, revealing a superiority of the Convolutional Neural Network (CNN), which, despite requiring greater computational cost and training time, provided high precision in the classifications and valuable insights through the application of the LIME explainability technique. Moreover, the study proposes a majority vote classification methodology for frequently observed events, enabling reliable classifications through models with moderate performance. It is concluded that the choice of the ideal model for the classification of soundscapes of the Cerrado should consider a balance between accuracy, operational complexity, and efficiency. The conclusions of this study offer relevant directions for future research and the application of monitoring technologies in conservation and recovery efforts of biomes.

The maned wolf (Chrysocyon brachyurus; Illiger, 1815) is considered the largest canid in South America, it has crepuscular nocturnal behavior, solitary and skittish habits. Its vocalization has different classifications, one of them called roar-barks, which helps in long-distance communication, being ideal for passive acoustic monitoring. For the study of the vocalizations of C. brachyurus in the Ecological Station of Itirapina in Brazil, were implanted 6 autonomous recorders called EcoPods. The locations of the station with the highest detection of vocalizations, their characteristics, temporal distribution in relation to sunset and moon phases were studied. The point with the highest concentration of roar-barks is the same as the point with the litter’s vocalization, indicating the importance of vocalizations in parental care. Passive acoustic monitoring allows a less invasive study and helps to understand the behavior of the maned wolf for its preservation.

Acoustic communication, broadly distributed along the vertebrate phylogeny, plays a fundamental role in parental care, mate attraction and various other behaviours. Despite its importance, comparatively less is known about the evolutionary roots of acoustic communication. Phylogenetic comparative analyses can provide insights into the deep time evolutionary origin of acoustic communication, but they are often plagued by missing data from key species. Here we present evidence for 53 species of four major clades (turtles, tuatara, caecilian and lungfish) in the form of vocal recordings and contextual behavioural information accompanying sound production. This and a broad literature-based dataset evidence acoustic abilities in several groups previously considered non-vocal. Critically, phylogenetic analyses encompassing 1800 species of choanate vertebrates reconstructs acoustic communication as a homologous trait, and suggests that it is at least as old as the last common ancestor of all choanate vertebrates, that lived approx. 407 million years before present.

Avian survey is a time-consuming and challenging task, often being conducted in remote and sometimes inhospitable locations. In this context, the development of automated acoustic landscape monitoring systems for bird survey is essential. We conducted a comparative study between two machine learning methods for the detection and identification of 2 endangered Brazilian bird species from the Psittacidae species, the Amazona brasiliensis and the Amazona vinacea . Specifically, we focus on the identification of these 2 species in an acoustic landscape where similar vocalizations from other Psittacidae species are present. A 3-step approach is presented, composed of signal segmentation and filtering, feature extraction, and classification. In the feature extraction step, the Mel-Frequency Cepstrum Coefficients features were extract and fed to the Random Forest Algorithm and the Multilayer Perceptron for training and classifying acoustic samples. The experiments showed promising results, particularly for the Random Forest algorithm, achieving accuracy of up to 99%. Using a combination of signal segmentation and filtering before the feature extraction steps greatly increased experimental results. Additionally, the results show that the proposed approach is robust and flexible to be adopted in passive acoustic monitoring systems.

Changepoint analysis (also known as segmentation analysis) aims to analyze an ordered, one-dimensional vector in order to find locations where some characteristic of the data changes. Many models and algorithms have been studied under this theme, including models for changes in mean and/or variance, changes in linear regression parameters, etc. This work is interested in an algorithm for the segmentation of long duration acoustic signals; the segmentation is based on the change of the root-mean-square power of the signal. It investigates a Bayesian model with two possible parameterizations and proposes a binary algorithm in two versions using non-informative or informative priors. These algorithms are tested in the segmentation of annotated acoustic signals from the Alcatrazes marine preservation park in Brazil.

Sound emissions by ships and boats can strongly impact marine life, with potential to affect communications, breeding and prey and predator relationships. Automatic detection of boat signatures in underwater audio recordings is thus an important task. Automated solutions are particularly relevant for monitoring preservation areas where the presence of watercrafts is usually regulated. The task is particularly challenging because it requires distinguishing multiple overlapping acoustic events in typically noisy audio recordings. In this paper, we introduce an algorithm for boat and ship detection which computes an acoustic signature that captures the variance in the frequency amplitudes observed over the duration of the signal. We evaluated the algorithm on a database of underwater recordings collected at two conservation areas in the State of São Paulo, Brazil, with very good results, and also compared it with an existing solution. Besides being effective, the algorithm requires limited user input and no parameter fine tuning to handle diverse situations. It thus provides a solution to automate the detection of vessels, with potential applications for monitoring marine preservation areas.

El ruido magnético Barkhausen (RMB) es un fenómeno que ocurre en materiales ferromagnéticos cuando se someten a campos magnéticos lentamente variables, generándose cambios en su magnetización, produciéndose “saltos” discretos. Estos saltos se generan por el movimiento de las paredes de los dominios magnéticos. Se detectan colocando una pequeña bobina sobre la superficie de la muestra. El movimiento de las paredes es regulado por la microestructura del material (precipitados, tamaño de grano, tensiones residuales, etc.). Es esta manera indirecta la que permite caracterizar al material. En este trabajo se presentan las mediciones del RMB para estudiar la anisotropía magnética, inducida por el proceso de laminación, y los cambios producidos por deformación plástica uniaxial en probetas de aceros inoxidables ferríticos AISI: 409, 430, 441A. Las mediciones se realizaron a diferentes porcentajes de deformación, con el fin de observar cómo cambian las características de las señales de RMB. Se presentan también los estudios métalográficos y de dureza. Los resultados muestran una clara tendencia en relación con la dirección de los esfuerzos principales.