Unsupervised classification to improve the quality of a bird song recording dataset

Open audio databases such as Xeno-Canto are widely used to build datasets to explore bird song repertoire or to train models for automatic bird sound classification by deep learning algorithms. However, such databases suffer from the fact that bird sounds are weakly labelled: a species name is attributed to each audio recording without timestamps that provide the temporal localization of the bird song of interest. Manual annotations can solve this issue, but they are time consuming, expert-dependent, and cannot run on large datasets. Another solution consists in using a labelling function that automatically segments audio recordings before assigning a label to each segmented audio sample. Although labelling functions were introduced to expedite strong label assignment, their classification performance remains mostly unknown. To address this issue and reduce label noise (wrong label assignment) in large bird song datasets, we introduce a data-centric novel labelling function composed of three successive steps: 1) time-frequency sound unit segmentation, 2) feature computation for each sound unit, and 3) classification of each sound unit as bird song or noise with either an unsupervised DBSCAN algorithm or the supervised BirdNET neural network. The labelling function was optimized, validated, and tested on the songs of 44 West-Palearctic common bird species. We first showed that the segmentation of bird songs alone aggregated from 10% to 83% of label noise depending on the species. We also demonstrated that our labelling function was able to significantly reduce the initial label noise present in the dataset by up to a factor of three. Finally, we discuss different opportunities to design suitable labelling functions to build high-quality animal vocalizations with minimum expert annotation effort.     

Passive acoustic monitoring of animal populations with transfer learning

Progress in deep learning, more specifically in using convolutional neural networks (CNNs) for the creation of classification models, has been tremendous in recent years. Within bioacoustics research, there has been a large number of recent studies that use CNNs. Designing CNN architectures from scratch is non-trivial and requires knowledge of machine learning. Furthermore, hyper-parameter tuning associated with CNNs is extremely time consuming and requires expensive hardware. In this paper we assess whether it is possible to build good bioacoustic classifiers by adapting and re-using existing CNNs pre-trained on the ImageNet dataset – instead of designing them from scratch, a strategy known as transfer learning that has proved highly successful in other domains. This study is a first attempt to conduct a large-scale investigation on how transfer learning can be used for passive acoustic monitoring (PAM), to simplify the implementation of CNNs and the design decisions when creating them, and to remove time consuming hyper-parameter tuning phases. We compare 12 modern CNN architectures across 4 passive acoustic datasets that target calls of the Hainan gibbon Nomascus hainanus, the critically endangered black-and-white ruffed lemur Varecia variegata, the vulnerable Thyolo alethe Chamaetylas choloensis, and the Pin-tailed whydah Vidua macroura. We focus our work on data scarcity issues by training PAM binary classification models very small datasets, with as few as 25 verified examples. Our findings reveal that transfer learning can result in up to 82% F1 score while keeping CNN implementation details to a minimum, thus rendering this approach accessible, easier to design, and speeding up further vocalisation annotations to create PAM robust models.     

Ecological inferences about marine mammals from passive acoustic data

Monitoring on the basis of sound recordings, or passive acoustic monitoring, can complement or serve as an alternative to real-time visual or aural monitoring of marine mammals and other animals by human observers. Passive acoustic data can support the estimation of common, individual-level ecological metrics, such as presence, detection-weighted occupancy, abundance and density, population viability and structure, and behaviour. Passive acoustic data also can support estimation of some community-level metrics, such as species richness and composition. The feasibility of estimation and certainty of estimates is highly context dependent, and understanding the factors that affect the reliability of measurements is useful for those considering whether to use passive acoustic data. Here, we review basic concepts and methods of passive acoustic sampling in marine systems that often are applicable to marine mammal research and conservation. Our ultimate aim is to facilitate collaboration among ecologists, bioacousticians, and data analysts. Ecological applications of passive acoustics require one to make decisions about sampling design, which in turn requires consideration of sound propagation, sampling of signals, and data storage. One also must make decisions about signal detection and classification and evaluation of the performance of algorithms for these tasks. Investment in the research and development of systems that automate detection and classification, including machine learning, are increasing. Passive acoustic monitoring is more reliable for detection of species presence than for estimation of other species-level metrics. Use of passive acoustic monitoring to distinguish among individual animals remains difficult. However, information about detection probability, vocalisation or cue rate, and relations between vocalisations and the number and behaviour of animals increases the feasibility of estimating abundance or density. Most sensor deployments are fixed in space or are sporadic, making temporal turnover in species composition more tractable to estimate than spatial turnover. Collaborations between acousticians and ecologists are most likely to be successful and rewarding when all partners critically examine and share a fundamental understanding of the target variables, sampling process, and analytical methods.     

Deep learning representations to support COVID-19 diagnosis on CT slices

Introduction: The coronavirus disease 2019 (COVID-19) has become a significant public health problem worldwide. In this context, CT-scan automatic analysis has emerged as a COVID-19 complementary diagnosis tool allowing for radiological finding characterization, patient categorization, and disease follow-up. However, this analysis depends on the radiologist’s expertise, which may result in subjective evaluations.Objective: To explore deep learning representations, trained from thoracic CT-slices, to automatically distinguish COVID-19 disease from control samples.Materials and methods: Two datasets were used: SARS-CoV-2 CT Scan (Set-1) and FOSCAL clinic’s dataset (Set-2). The deep representations took advantage of supervised learning models previously trained on the natural image domain, which were adjusted following a transfer learning scheme. The deep classification was carried out: (a) via an end-to-end deep learning approach and (b) via random forest and support vector machine classifiers by feeding the deep representation embedding vectors into these classifiers.Results: The end-to-end classification achieved an average accuracy of 92.33% (89.70% precision) for Set-1 and 96.99% (96.62% precision) for Set-2. The deep feature embedding with a support vector machine achieved an average accuracy of 91.40% (95.77% precision) and 96.00% (94.74% precision) for Set-1 and Set-2, respectively.Conclusion: Deep representations have achieved outstanding performance in the identification of COVID-19 cases on CT scans demonstrating good characterization of the COVID-19 radiological patterns. These representations could potentially support the COVID-19 diagnosis in clinical settings.     

Hierarchical Gene Selection and Genetic Fuzzy System for Cancer Microarray Data Classification

This paper introduces a novel approach to gene selection based on a substantial modification of analytic hierarchy process (AHP). The modified AHP systematically integrates outcomes of individual filter methods to select the most informative genes for microarray classification. Five individual ranking methods including t-test, entropy, receiver operating characteristic (ROC) curve, Wilcoxon and signal to noise ratio are employed to rank genes. These ranked genes are then considered as inputs for the modified AHP. Additionally, a method that uses fuzzy standard additive model (FSAM) for cancer classification based on genes selected by AHP is also proposed in this paper. Traditional FSAM learning is a hybrid process comprising unsupervised structure learning and supervised parameter tuning. Genetic algorithm (GA) is incorporated in-between unsupervised and supervised training to optimize the number of fuzzy rules. The integration of GA enables FSAM to deal with the high-dimensional-low-sample nature of microarray data and thus enhance the efficiency of the classification. Experiments are carried out on numerous microarray datasets. Results demonstrate the performance dominance of the AHP-based gene selection against the single ranking methods. Furthermore, the combination of AHP-FSAM shows a great accuracy in microarray data classification compared to various competing classifiers. The proposed approach therefore is useful for medical practitioners and clinicians as a decision support system that can be implemented in the real medical practice.     

Contributions of MIR to soundscape ecology. Part 3: Tagging and classifying audio features using a multi-labeling k-nearest neighbor approach

Scientists are using acoustic monitoring to assess the impact of altered soundscapes on wildlife communities and human systems. In the soundscape ecology field, monitoring and analyses approaches rely on the interdisciplinary intersection of ecology, acoustics, and computer science. Combining theory and practice of each field in the context of Knowledge Discovery in Databases (KDD), soundscape ecologists provide innovative monitoring solutions for ecologically-driven research questions. We propose a soundscape content analysis framework for improved knowledge outcome with assistance of the new multi-label (ML) concept.Here, we investigated the effectiveness of a ML k-nearest neighbor algorithm (ML-kNN) for labeling concurrent soundscape components within a single recording. We manually labeled 1200 field recordings for the presence of soundscape components and extracted ecological acoustic features, audio profile features, and Gaussian-mixture model features for each recording. Then, we tested the ML-kNN algorithm accuracy with well-established metrics adapted to ML learning.We found that seventeen unique acoustic features could predict a set of biophonic, geophonic, and anthrophonic labels for a single field recording with average precision of 0.767. However, certain labels were predicted incorrectly depending on the time of day and co-occurrence of that label with another label, suggesting further refinement is needed to improve the accuracy of predicted labels.Overall, this ML classification approach could enable researchers to label field recordings more quickly and generate an “alert” system for monitoring changes in a specific sound class. Ultimately, the adaptation of the ML algorithm may provide soundscape ecologists with new metadata labels that are searchable in large databases of soundscape field recordings.     

Machine learning monitoring for laser osteotomy

This work proposes a new online monitoring method for an assistance during laser osteotomy. The method allows differentiating the type of ablated tissue and the applied dose of laser energy. The setup analyzes the laser-induced acoustic emission, detected by an airborne microphone sensor. The analysis of the acoustic signals is carried out using a machine learning algorithm that is pre-trained in a supervised manner. The efficiency of the method is experimentally evaluated with several types of tissues, which are: skin, fat, muscle, and bone. Several cutting-edge machine learning frameworks are tested for the comparison with the resulting classification accuracy in the range of 84–99%. It is shown that the datasets for the training of the machine learning algorithms are easy to collect in real-life conditions. In the future, this method could assist the doctors during laser osteotomy, minimizing the damage of the nearby healthy tissues and provide cleaner pathologic tissue removal.     

Using machine learning to predict the effects and consequences of mutations in proteins

Machine and deep learning approaches can leverage the increasingly available massive datasets of protein sequences, structures, and mutational effects to predict variants with improved fitness. Many different approaches are being developed, but systematic benchmarking studies indicate that even though the specifics of the machine learning algorithms matter, the more important constraint comes from the data availability and quality utilized during training. In cases where little experimental data are available, unsupervised and self-supervised pre-training with generic protein datasets can still perform well after subsequent refinement via hybrid or transfer learning approaches. Overall, recent progress in this field has been staggering, and machine learning approaches will likely play a major role in future breakthroughs in protein biochemistry and engineering.     

Probing real sensory worlds of receivers with unsupervised clustering

The task of an organism to extract information about the external environment from sensory signals is based entirely on the analysis of ongoing afferent spike activity provided by the sense organs. We investigate the processing of auditory stimuli by an acoustic interneuron of insects. In contrast to most previous work we do this by using stimuli and neurophysiological recordings directly in the nocturnal tropical rainforest, where the insect communicates. Different from typical recordings in sound proof laboratories, strong environmental noise from multiple sound sources interferes with the perception of acoustic signals in these realistic scenarios. We apply a recently developed unsupervised machine learning algorithm based on probabilistic inference to find frequently occurring firing patterns in the response of the acoustic interneuron. We can thus ask how much information the central nervous system of the receiver can extract from bursts without ever being told which type and which variants of bursts are characteristic for particular stimuli. Our results show that the reliability of burst coding in the time domain is so high that identical stimuli lead to extremely similar spike pattern responses, even for different preparations on different dates, and even if one of the preparations is recorded outdoors and the other one in the sound proof lab. Simultaneous recordings in two preparations exposed to the same acoustic environment reveal that characteristics of burst patterns are largely preserved among individuals of the same species. Our study shows that burst coding can provide a reliable mechanism for acoustic insects to classify and discriminate signals under very noisy real-world conditions. This gives new insights into the neural mechanisms potentially used by bushcrickets to discriminate conspecific songs from sounds of predators in similar carrier frequency bands.     

Cryo-EM image alignment: From pair-wise to joint with deep unsupervised difference learning

Cryo-electron microscopy (cryo-EM) single-particle analysis is a revolutionary imaging technique to resolve and visualize biomacromolecules. Image alignment in cryo-EM is an important and basic step to improve the precision of the image distance calculation. However, it is a very challenging task due to high noise and low signal-to-noise ratio. Therefore, we propose a new deep unsupervised difference learning (UDL) strategy with novel pseudo-label guided learning network architecture and apply it to pair-wise image alignment in cryo-EM. The training framework is fully unsupervised. Furthermore, a variant of UDL called joint UDL (JUDL), is also proposed, which is capable of utilizing the similarity information of the whole dataset and thus further increase the alignment precision. Assessments on both real-world and synthetic cryo-EM single-particle image datasets suggest the new unsupervised joint alignment method can achieve more accurate alignment results. Our method is highly efficient by taking advantages of GPU devices. The source code of our methods is publicly available at “http://www.csbio.sjtu.edu.cn/bioinf/JointUDL/” for academic use.     

Assessing marine ecosystem acoustic diversity across ocean basins

Concurrent with the elevation of the concern over the state of sound in the ocean, advances in terrestrial acoustic monitoring techniques have produced concepts and tools that may be applicable to the underwater world. Several index values that convey information related to acoustic diversity with a single numeric measurement made from acoustic recordings have been proposed for rapidly assessing community biodiversity. Here we apply the acoustic biodiversity index method to low frequency recordings made from three different ocean basins to assess its appropriateness for characterizing species richness in the marine environment. Initial results indicated that raw acoustic entropy (H) values did not correspond to biological patterns identified from individual signal detections and classification. Noise from seismic airgun activity masked the weaker biological signals and confounded the entropy calculation. A simple background removal technique that subtracted an average complex spectrum characteristic of seismic exploration signals from the average spectra of each analysis period that contained seismic signals was applied to compensate for salient seismic airgun signals present in all locations. The noise compensated (H_N) entropy index was more reflective of biological patterns and holds promise for the use of rapid acoustic biodiversity in the marine environment as an indicator of habitat biodiversity and health.     

LEDPatNet19: Automated Emotion Recognition Model based on Nonlinear LED Pattern Feature Extraction Function using EEG Signals

Electroencephalography (EEG) signals collected from human brains have generally been used to diagnose diseases. Moreover, EEG signals can be used in several areas such as emotion recognition, driving fatigue detection. This work presents a new emotion recognition model by using EEG signals. The primary aim of this model is to present a highly accurate emotion recognition framework by using both a hand-crafted feature generation and a deep classifier. The presented framework uses a multilevel fused feature generation network. This network has three primary phases, which are tunable Q-factor wavelet transform (TQWT), statistical feature generation, and nonlinear textural feature generation phases. TQWT is applied to the EEG data for decomposing signals into different sub-bands and create a multilevel feature generation network. In the nonlinear feature generation, an S-box of the LED block cipher is utilized to create a pattern, which is named as Led-Pattern. Moreover, statistical feature extraction is processed using the widely used statistical moments. The proposed LED pattern and statistical feature extraction functions are applied to 18 TQWT sub-bands and an original EEG signal. Therefore, the proposed hand-crafted learning model is named LEDPatNet19. To select the most informative features, ReliefF and iterative Chi2 (RFIChi2) feature selector is deployed. The proposed model has been developed on the two EEG emotion datasets, which are GAMEEMO and DREAMER datasets. Our proposed hand-crafted learning network achieved 94.58%, 92.86%, and 94.44% classification accuracies for arousal, dominance, and valance cases of the DREAMER dataset. Furthermore, the best classification accuracy of the proposed model for the GAMEEMO dataset is equal to 99.29%. These results clearly illustrate the success of the proposed LEDPatNet19.     

Automated classification of bat echolocation call recordings with artificial intelligence

Acoustic recorders are commonly used to remotely monitor and collect data on bats (Order Chiroptera). These efforts result in many acoustic recordings that must be classified by a bat biologist with expertise in call classification in order to obtain useful information. The rarity of this expertise and time constraints have prompted efforts to automatically classify bat species in acoustic recordings using a variety of learning methods. There are several software programs available for this purpose, but they are imperfect and the United States Fish and Wildlife Service often recommends that a qualified acoustic analyst review bat call identifications even if using these software programs. We sought to build a model to classify bat species using modern computer vision techniques. We used images of bat echolocation calls (i.e., plots of the pulses) to train deep learning computer vision models that automatically classify bat calls to species. Our model classifies 10 species, five of which are protected under the Endangered Species Act. We evaluated our models using standard model validation procedures, and performed two external tests. For these tests, an entire dataset was withheld from the procedure before splitting the data into training and validation sets. We found that our validation accuracy (92%) and testing accuracy (90%) were higher than when we used Kaleidoscope Pro and BCID software (65% and 61% accuracy, respectively) to evaluate the same calls. Our results suggest that our approach is effective at classifying bat species from acoustic recordings, and our trained model will be incorporated into new bat call identification software: WEST-EchoVision.     

Multi-category classification using an Extreme Learning Machine for microarray gene expression cancer diagnosis

In this paper, the recently developed Extreme Learning Machine (ELM) is used for direct multicategory classification problems in the cancer diagnosis area. ELM avoids problems like local minima, improper learning rate and overfitting commonly faced by iterative learning methods and completes the training very fast. We have evaluated the multi-category classification performance of ELM on three benchmark microarray datasets for cancer diagnosis, namely, the GCM dataset, the Lung dataset and the Lymphoma dataset. The results indicate that ELM produces comparable or better classification accuracies with reduced training time and implementation complexity compared to artificial neural networks methods like conventional back-propagation ANN, Linder's SANN, and Support Vector Machine methods like SVM-OVO and Ramaswamy's SVM-OVA. ELM also achieves better accuracies for classification of individual categories.     

Echolocation by the barbastelle bat, Barbastella barbastellus

When searching for insects along edges, Barbastella barbastellus alternated between two signal types. Type-2 signals had durations around 6 ms and were composed of an initial shallowly downward frequency modulated component, starting at about 45 kHz and followed by a shorter more steeply modulated component that ended at about 32 kHz. Type-1 signals were rather stereotyped with durations around 2.5 ms and a very short rise time. They covered an approximately 8 kHz-wide frequency band positioned just below the 12-15 kHz-wide frequency band of type-2 signals, with no or small frequency overlap. In the recordings, type-1 signals almost had always a higher amplitude than type-2 signals, at least partly caused by head movements. Assuming that signal structure reflects function, we hypothesize that type-2 signals have the same adaptive value as the signals with a broadband and narrowband component of other vespertilionids, but with a reverse arrangement of the signal elements. Like the broadband component of the type-2 signals, type-1 signals are well suited to localize background targets. Thus, the localization component may be distributed among two signals separated in time, which has the advantage that both signals can be varied independently in the direction of emission and in amplitude.     

The effect of soundscape composition on bird vocalization classification in a citizen science biodiversity monitoring project

There is a need for monitoring biodiversity at multiple spatial and temporal scales to aid conservation efforts. Autonomous recording units (ARUs) can provide cost-effective, long-term and systematic species monitoring data for sound-producing wildlife, including birds, amphibians, insects and mammals over large areas. Modern deep learning can efficiently automate the detection of species occurrences in these sound data with high accuracy. Further, citizen science can be leveraged to scale up the deployment of ARUs and collect reference vocalizations needed for training and validating deep learning models. In this study we develop a convolutional neural network (CNN) acoustic classification pipeline for detecting 54 bird species in Sonoma County, California USA, with sound and reference vocalization data collected by citizen scientists within the Soundscapes to Landscapes project (www.soundscapes2landscapes.org). We trained three ImageNet-based CNN architectures (MobileNetv2, ResNet50v2, ResNet100v2), which function as a Mixture of Experts (MoE), to evaluate the usefulness of several methods to enhance model accuracy. Specifically, we: 1) quantify accuracy with fully-labeled 1-min soundscapes for an assessment of real-world conditions; 2) assess the effect on precision and recall of additional pre-training with an external sound archive (xeno-canto) prior to fine-tuning with vocalization data from our study domain; and, 3) assess how detections and errors are influenced by the presence of coincident biotic and non-biotic sounds (i.e., soundscape components). In evaluating accuracy with soundscape data (n = 37 species) across CNN probability thresholds and models, we found acoustic pre-training followed by fine-tuning improved average precision by 10.3% relative to no pre-training, although there was a small average 0.8% reduction in recall. In selecting an optimal CNN architecture for each species based on maximum F(β = 0.5), we found our MoE approach had total precision of 84.5% and average species precision of 85.1%. Our data exhibit multiple issues arising from applying citizen science and acoustic monitoring at the county scale, including deployment of ARUs with relatively low fidelity and recordings with background noise and overlapping vocalizations. In particular, human noise was significantly associated with more incorrect species detections (false positives, decreased precision), while physical interference (e.g., recorder hit by a branch) and geophony (e.g., wind) was associated with the classifier missing detections (false negatives, decreased recall). Our process surmounted these obstacles, and our final predictions allowed us to demonstrate how deep learning applied to acoustic data from low-cost ARUs paired with citizen science can provide valuable bird diversity data for monitoring and conservation efforts.     

Automated auditory detection of a rare,secretive marsh bird with infrequent and acoustically indistinct vocalizations

Autonomous recording units (ARUs) provide a non-invasive and efficient method for acoustic detection of elusive species across large temporal and spatial scales. However, species with indistinct vocalization structures can be a considerable challenge for automated signal recognizers. We investigated the performance of ARUs and signal recognizers in identifying the broadband, short-syllable, pulsed calls of a secretive, threatened marsh bird, the King Rail Rallus elegans. Other sympatric species in the same habitat also have repetitive calls within the same frequency range that can be difficult to distinguish. Following serial ARU deployments at specified sites in known breeding habitat, we conducted standardized callback surveys and nest searches to provide an independent measure of breeder density. To analyse recordings, we developed a signal recognizer based on user-input training files to detect two common call types, kek and grunt. Detections that remained following manual review of recognizer output revealed a previously undescribed seasonal decline and crepuscular diel pattern in calling rate. The rate of the grunt call also predicted density. These patterns emerged despite the recognizer's low precision and high false-positive rate, which were largely due to misclassification of other species' calls, although ambient noise and effective detection radius also limited the detectability of King Rail calls. We demonstrate that with informed ARU scheduling, improved ability to manipulate user-specified parameters within signal detection software, and attention to quality control, even the simplest call structures can be located consistently in a diverse acoustic landscape. Our behavioural findings will inform improvements to auditory surveys and to management of King Rails across their range.     

Using supervised fuzzy clustering to predict protein structural classes

Prediction of protein classification is both an important and a tempting topic in protein science. This is because of not only that the knowledge thus obtained can provide useful information about the overall structure of a query protein, but also that the practice itself can technically stimulate the development of novel predictors that may be straightforwardly applied to many other relevant areas. In this paper, a novel approach, the so-called "supervised fuzzy clustering approach" is introduced that is featured by utilizing the class label information during the training process. Based on such an approach, a set of "if-then" fuzzy rules for predicting the protein structural classes are extracted from a training dataset. It has been demonstrated through two different working datasets that the overall success prediction rates obtained by the supervised fuzzy clustering approach are all higher than those by the unsupervised fuzzy c-means introduced by the previous investigators [C.T. Zhang, K.C. Chou, G.M. Maggiora. Protein Eng. (1995) 8, 425-435]. It is anticipated that the current predictor may play an important complementary role to other existing predictors in this area to further strengthen the power in predicting the structural classes of proteins and their other characteristic attributes.     

A stable iterative method for refining discriminative gene clusters

Background

Microarray technology is often used to identify the genes that are differentially expressed between two biological conditions. On the other hand, since microarray datasets contain a small number of samples and a large number of genes, it is usually desirable to identify small gene subsets with distinct pattern between sample classes. Such gene subsets are highly discriminative in phenotype classification because of their tightly coupling features. Unfortunately, such identified classifiers usually tend to have poor generalization properties on the test samples due to overfitting problem.

Results

We propose a novel approach combining both supervised learning with unsupervised learning techniques to generate increasingly discriminative gene clusters in an iterative manner. Our experiments on both simulated and real datasets show that our method can produce a series of robust gene clusters with good classification performance compared with existing approaches.

Conclusion

This backward approach for refining a series of highly discriminative gene clusters for classification purpose proves to be very consistent and stable when applied to various types of training samples.