Juan Gabriel Colonna

An incremental technique for real-time bioacoustic signal segmentation

An incremental transformation of ZCR and energy without using temporal windows.With our method is possible to save memory and transmission costs.Solution to process large amounts of data by resource-constrained devices as WSN. A bioacoustical animal recognition system is composed of two parts: (1) the segmenter, responsible for detecting syllables (animal vocalization) in the audio; and (2) the classifier, which determines the species/animal whose the syllables belong to. In this work, we first present a novel technique for automatic segmentation of anuran calls in real time; then, we present a method to assess the performance of the whole system. The proposed segmentation method performs an unsupervised binary classification of time series (audio) that incrementally computes two exponentially-weighted features (Energy and Zero Crossing Rate). In our proposal, classical sliding temporal windows are replaced with counters that give higher weights to new data, allowing us to distinguish between a syllable and ambient noise (considered as silences). Compared to sliding-window approaches, the associated memory cost of our proposal is lower, and processing speed is higher. Our evaluation of the segmentation component considers three metrics: (1) the Matthews Correlation Coefficient for point-to-point comparison; (2) the WinPR to quantify the precision of boundaries; and (3) the AEER for event-to-event counting. The experiments were carried out in a dataset with 896 syllables of seven different species of anurans. To evaluate the whole system, we derived four equations that helps understand the impact that the precision and recall of the segmentation component has on the classification task. Finally, our experiments show a segmentation/recognition improvement of 37%, while reducing memory and data communication. Therefore, results suggest that our proposal is suitable for resource-constrained systems, such as Wireless Sensor Networks (WSNs).

Similarity clustering for data fusion in Wireless Sensor Networks using k-means

Wireless Sensor Networks consist of a powerful technology for monitoring the physical world. Particularly, in-network data fusion techniques are very important to applications such as target classification and tracking to reduce the communication burden in these constrained networks. However, the efficiency of the solution can be affected by the data correlation among several sensor nodes. Thus, the application of value fusion (for clusters of nodes with correlated measurements) and decision fusion (combining the local decisions of the clusters) is a common strategy. In this work, we propose an algorithm for properly selecting the groups of nodes with correlated measurements. Experiments show that our algorithm is 30% better than a solution that considers only the spatial coherence regions.

Feature subset selection for automatically classifying anuran calls using sensor networks

Anurans (frogs or toads) are commonly used by biologists as early indicators of ecological stress. The reason is that anurans are closely related to the ecosystem. Although several sources of data may be used for monitoring these animals, anuran calls lead to a non-intrusive data acquisition strategy. Moreover, wireless sensor networks (WSNs) may be used for such a task, resulting in more accurate and autonomous system. However, it is essential save resources to extend the network lifetime. In this paper, we evaluate the impact of reducing data dimension for automatic classification of bioacoustic signals when a WSN is involved. Such a reduction is achieved through a wrapper-based feature subset selection strategy that uses genetic algorithm (GA). We use GA to find the subset of features that maximizes the cost-benefit ratio. In addition, we evaluate the impact of reducing the original feature space, when sampling frequencies are also reduced. Experimental results indicate that we can reduce the number of features, while increasing classification rates (even when smaller sampling frequencies of transmission are used).

Automatic Classification of Anuran Sounds Using Convolutional Neural Networks

Anurans (frogs or toads) are closely related to the ecosystem and they are commonly used by biologists as early indicators of ecological stress. Automatic classification of anurans, by processing their calls, helps biologists analyze the activity of anurans on larger scale. Wireless Sensor Networks (WSNs) can be used for gathering data automatically over a large area. WSNs usually set restrictions on computing and transmission power for extending the networks lifetime. Deep Learning algorithms have gathered a lot of popularity in recent years, especially in the field of image recognition. Being an eager learner, a trained Deep Learning model does not need a lot of computing power and could be used in hardware with limited resources. This paper investigates the possibility of using Convolutional Neural Networks with Mel-Frequency Cepstral Coefficients (MFCCs) as input for the task of classifying anuran sounds.

A Distributed Approach for Classifying Anuran Species Based on Their Calls

In this work, we evaluate the performance of a distributed classification system in a Wireless Sensor Network for monitoring anurans. Our aim is to study how to take advantage of the collaborative nature of the sensor network to improve the recognition of anuran calls. To accomplish this, we evaluate four low-cost techniques (majority vote, weighted majority vote, arithmetic and geometric combinators) to combine three classifiers commonly used in sensor applications (Quadratic Discriminant Analysis, Naive Bayes, and Decision Trees) and trained to identify anuran calls. We investigate how the environment perceptions of the sensors can be used to discard confusing scenarios, i.e., scenarios in which there are multiple calls from different species at same time. Our best combination strategy achieved a gain of about 11% over a sensor taken in isolation. We also found that, by using the entropy of the species estimates, the sensor committee is able to effectively identify confusing scenarios, increasing gains over the isolated sensor to about 20%.

How to Correctly Evaluate an Automatic Bioacoustics Classification Method

In this work, we introduce a more appropriate (or alternative) approach to evaluate the performance and the generalization capabilities of a framework for automatic anuran call recognition. We show that, by using the common k-folds Cross-Validation (k-CV) procedure to evaluate the expected error in a syllable-based recognition system the recognition accuracy is overestimated. To overcome this problem, and to provide a fair evaluation, we propose a new CV procedure in which the specimen information is considered during the split step of the k-CV. Therefore, we performed a k-CV by specimens (or individuals) showing that the accuracy of the system decrease considerably. By introducing the specimen information, we are able to answer a more fundamental question: Given a set of syllables that belongs to a specific group of individuals, can we recognize new specimens of the same species? In this article, we go deeper into the reviews and the experimental evaluations to answer this question.

Recognizing Family, Genus, and Species of Anuran Using a Hierarchical Classification Approach

In bioacoustic recognition approaches, a “flat” classifier is usually trained to recognize several species of anuran, where the number of classes is equal to the number of species. Consequently, the complexity of the classification function increases proportionally to the amount of species. To avoid this issue we propose a “hierarchical” approach that decomposes the problem into three taxonomic levels: the family, the genus, and the species level. To accomplish this, we transform the original single-label problem into a multi-dimensional problem (multi-label and multi-class) considering the Linnaeus taxonomy. Then, we develop a top-down method using a set of classifiers organized as a hierarchical tree. Thus, it is possible to predict the same set of species as a flat classifier, and additionally obtain new information about the samples and their taxonomic relationship. This helps us to understand the problem better and achieve additional conclusions by the inspection of the confusion matrices at the three levels of classification. In addition, we carry out our experiments using a Cross-Validation performed by individuals. This form of CV avoids mixing syllables that belong to the same specimens in the testing and training sets, preventing an overestimate of the accuracy and generalizing the predictive capabilities of the system. We tested our system in a dataset with sixty individual frogs, from ten different species, eight genus, and four families, achieving a final Micro- and Average-accuracy equal to 86 % and 62 % respectively.

Compressive Sensing for Efficiently Collecting Wildlife Sounds with Wireless Sensor Networks

Wildlife sounds provide relevant information for non-intrusive environmental monitoring when Wireless Sensor Networks (WSNs) are used. Thus, collecting such audio data, while maximizing the network lifetime, is a key challenge for WSNs. In this work, we propose a methodology that applies Compressive Sensing (CS) aiming at collecting as little data as possible to allow the signal reconstruction, so that the reconstructed signal is still representative. The key issue is to determine a sparse base that best represents the audio information used for identifying the target species. As a proof-of-concept, we focus on anuran (frogs and toads) calls, but the methodology can be applied for other animal families and species. The reason for that choice is that long-term anuran monitoring has been used by biologists as an early indicator for ecological stress. By using real wild anuran calls, we show that 98% classification rate can be achieved by using as little as 10% of the original data. We also use simulation to evaluate the impact of our solution on the network performance (energy consumption, delivery rate, and network delay).

Feature evaluation for unsupervised bioacoustic signal segmentation of anuran calls

Abstract We present a comprehensive study of temporal Low-Level acoustic Descriptors (LLDs) to automatically segment anuran calls in audio streams. The acoustic segmentation, or syllable extraction, is a key task shared by most of the bioacoustical species recognition systems. Consequently, the syllable extraction has a direct impact on the classification rate. In this work, we assess several new entropy measures including the recently developed Permutation Entropy, Weighted Permutation Entropy, and Permutation Min-Entropy, and compare them to the classical Energy, Zero Crossing Rate and Spectral Entropy. In addition, we propose an algorithm to estimate the optimal segmentation threshold value used to separate deterministic segments from stochastic ones avoiding the creation of thin clusters. To assess the performance of our segmentation approach, we applied a frame-by-frame, a point-to-point and an event-to-event comparisons. We show that in a scenario with severe noise conditions (SNR ≤ 0dB), simple entropy descriptors are robust, achieving 97% of segmentation performance, while keeping a low computational cost. We conclude that there is no LLD that is suitable for all scenarios, and we must adopt multiple or different LLDs, depending on the expected noise conditions.

Experimental Evaluation on Machine Learning Techniques for Human Activities Recognition in Digital Education Context

This paper describes an experimental evaluation of the main machine learning supervised techniques to be used for the human activities recognition in the context of technological education using data collected from smartphones sensors. The overall goal is to use the recognition of activities to identify students with attention deficit or hyperactivity problems, by recognizing three activities: walking, standing and sitting. Hence, this work focuses on developing activities recognition method of the students. The methodology consists in: collecting data where the user explicitly states what activity he/she is doing; applying various techniques to automatically recognize the activities; and measuring the degree of accuracy of each technique. The results shows that techniques such as Bayesian inference and SVM (Support Vector Machine) have smaller accuracy than techniques based on decision tree and kNN (k-nearest neighbors). Furthermore, the techniques based on decision trees have a constant computational cost, while the kNN depends on the number of samples.