Wildlife monitoring is often used by biologist to acquire information about animals and their habitat. In this context, animal sounds and vocalizations usually provide a specie fingerprint that is used for classifying the target species in a given site. For that matter, Wireless Sensor Networks (WSNs) represent an interesting option for automatically classifying animal species based on their vocalizations. In this work, we provide a solution that applies machine learning and signal processing techniques for classifying wildlife based on their vocalization. As a proof-of-concept, we choose anurans as the target animals. The reason is that anurans are already used by biologists as an early indicator of ecological stress, since they provide relevant information about terrestrian and aquatic ecosystems. Any solution must consider WSN limitations, trying to reduce the communication load to extend the network lifetime. Therefore, our solution represents the acoustic signals by a set of features. This representation allows us to identifiy specific signal patterns for each specie, reducing the amount of information necessary to classify it. Identifying such features, and/or combinations among them, is a key point to improve the solution benefit-cost ratio. As a consequence, we implemented and compared sets of existing features based on Fourier and Wavelet transforms. In our analysis, we first compare the sets of spectral and temporal characteristic, by using the entropy as a criterion for generating the combinations. Second, we reduce the set of features by using genetic algorithm. The proposed framework contains three steps: (i) the pre-processing to prepare the signals and perform the extraction of syllables, (ii) the extraction of features, and (iii) the species classification, using k-NN or SVM. Our experiments comprise four case studies, evaluating the effect of sampling frequency of the hardware and the number of bits used to represent each sample. This enable us to
conclude that, in enviromental monitoring using WSNs, the set of Mel coefficients is the most appropriate for classifying anuran calls.
Wildlife monitoring is often used by biologist to acquire information about animals and their habitat. In this context, animal sounds and vocalizations usually provide a specie fingerprint that is used for classifying the target species in a given site. For that matter, Wireless Sensor Networks (WSNs) represent an interesting option for automatically classifying animal species based on their vocalizations. In this work, we provide a solution that applies machine learning and signal processing techniques for classifying wildlife based on their vocalization. As a proof-of-concept, we choose anurans as the target animals. The reason is that anurans are already used by biologists as an early indicator of ecological stress, since they provide relevant information about terrestrian and aquatic ecosystems. Any solution must consider WSN limitations, trying to reduce the communication load to extend the network lifetime. Therefore, our solution represents the acoustic signals by a set of features. This representation allows us to identifiy specific signal patterns for each specie, reducing the amount of information necessary to classify it. Identifying such features, and/or combinations among them, is a key point to improve the solution benefit-cost ratio. As a consequence, we implemented and compared sets of existing features based on Fourier and Wavelet transforms. In our analysis, we first compare the sets of spectral and temporal characteristic, by using the entropy as a criterion for generating the combinations. Second, we reduce the set of features by using genetic algorithm. The proposed framework contains three steps: (i) the pre-processing to prepare the signals and perform the extraction of syllables, (ii) the extraction of features, and (iii) the species classification, using k-NN or SVM. Our experiments comprise four case studies, evaluating the effect of sampling frequency of the hardware and the number of bits used to represent each sample. This enable us to
conclude that, in enviromental monitoring using WSNs, the set of Mel coefficients is the most appropriate for classifying anuran calls.
Le cahier des charges du Forum des jeunes ambassadeurs de la francophonie des Amériques 2016 est l’outil à privilégier pour constituer votre proposition. Il a pour but de définir la nature et la structure de l’événement et de détailler les rôles et responsabilités d'un organisme hôte.
Le cahier des charges du Forum des jeunes ambassadeurs de la francophonie des Amériques 2016 est l’outil à privilégier pour constituer votre proposition. Il a pour but de définir la nature et la structure de l’événement et de détailler les rôles et responsabilités d'un organisme hôte.