Currently, video generation devices are simpler to manipulate, more portable and with lower prices. This allowed easy storage and transmission of large amounts of media, such as videos, which has facilitated the analysis of information, independent of human assistance for evaluation and exhaustive search of videos. Virtual reality, robotics, tele-medicine, humanmachine interface and tele-surveillance are applications for these techniques. This paper describes a method for human action recognition in videos using two convolutional neural networks (CNNs). The first one Spatial Stream (trained with frames of the video) and the second one Temporal Stream, trained with stacks of Dense Optical Flow (DOF). Both streams were trained separately and from both of them we generated a classification histogram based on the most frequent class assignment. For final classification, those histograms were combined to produce a single output. The technique was tested in two public action video datasets: Weizmann and UCF Sports. We achieve 84.44% of accuracy on Weizmann dataset for Spatial stream and 78.46% on UCF Sports dataset. For the Weizmann dataset we obtained 91.11% with networks combination.
An optoelectronic implementation of a modified FitzHugh–Nagumo neuron model is proposed, analyzed, and experimentally demonstrated. The setup uses linear optics and linear electronics for implementing an optical wavelength-domain nonlinearity. The system attains instability through a bifurcation mechanism present in a class of neuron models, a fact that is shown analytically. The implementation exhibits basic features of neural dynamics including threshold, production of short pulses (or spikes), and refractoriness.
This study aims to investigate the consensus among informants in the naming of tree species from a high diversity environment, the Brazilian Atlantic Forest (Sete Barras, SP), through a methodological procedure based on standardized stimuli. Seven selected local experts on tree species used for timber and handicrafts were asked to walk individually across the same area of 1.72 ha and identify and name all the known trees of more than 4 cm DBH (diameter at breast height) using common names. All trees were botanically identified, and their DBH and height were measured. The ecologic salience of tree species, expressed in terms of abundance, average height and DBH, was tested in relation to the informants' knowledge and species naming. The guided walks resulted on 708 identification events, with common names corresponding to 122 botanical species, or 68% of all tree species present. Both the reduced abundance and ecological salience of rare species can explain their recognition. The highest concordances in naming a tree were related only to the species abundance and not to their size (given by diameter and height). In some cases, there is no single common name for a botanical species, reflecting the intrinsic variation in local knowledge, which must be considered in ethnobotanical studies, in ecological assessments based on local knowledge, as well as in community-based conservation and management programs.
The operation of an optoelectronic dynamic neural model implementation is extended to higher frequencies. A simplified model of thermal effects in vertical-cavity surface-emitting lasers correctly predicts the qualitative changes in the nonlinear mapping implementation with frequency. Experiments and simulations show the expected resonance properties of this model neuron, along with the possibility of other dynamic effects in addition to the ones observed in the original FitzHugh–Nagumo equations. Results of optical coupling between two similar pulsing artificial neurons are also presented.
This paper presents a statistical analysis of stochastic-based optimization algorithms applied to a digital filter arbitrary magnitude and phase approximation design problem. Using an already developed rigorous statistical methodology, a completely randomized design is set up and best parameters values are estimated for the adaptive algorithms applied to a specific non-linear approximation problem. After finding the best parameter values, an additional completely randomized design is set up, comparing the performance of the adaptive algorithms with a quasi-Newton algorithm. Results for the statistical analysis are presented and the performance for different optimization algorithms with the best parameter values are analyzed.