Ricardo Tapiador-Morales

Performance Evaluation of Neural Networks for Animal Behaviors Classification: Horse Gaits Case Study

The study and monitoring of wildlife has always been a subject of great interest. Studying the behavior of wildlife animals is a very complex task due to the difficulties to track them and classify their behaviors through the collected sensory information. Novel technology allows designing low cost systems that facilitate these tasks. There are currently some commercial solutions to this problem; however, it is not possible to obtain a highly accurate classification due to the lack of gathered information. In this work, we propose an animal behavior recognition, classification and monitoring system based on a smart collar device provided with inertial sensors and a feed-forward neural network or Multi-Layer Perceptron (MLP) to classify the possible animal behavior based on the collected sensory information. Experimental results over horse gaits case study show that the recognition system achieves an accuracy of up to 95.6%.

System based on inertial sensors for behavioral monitoring of wildlife

Sensors Network is an integration of multiples sensors in a system to collect information about different environment variables. Monitoring systems allow us to determine the current state, to know its behavior and sometimes to predict what it is going to happen. This work presents a monitoring system for semi-wild animals that get their actions using an IMU (inertial measure unit) and a sensor fusion algorithm. Based on an ARM-CortexM4 microcontroller this system sends data using ZigBee technology of different sensor axis in two different operations modes: RAW (logging all information into a SD card) or RT (real-time operation). The sensor fusion algorithm improves both the precision and noise interferences.

Embedded neural network for real-time animal behavior classification

Recent biological studies have focused on understanding animal interactions and welfare. To help biologists to obtain animals behavior information, resources like wireless sensor networks are needed. Moreover, large amounts of obtained data have to be processed off-line in order to classify different behaviors. There are recent research projects focused on designing monitoring systems capable of measuring some animals parameters in order to recognize and monitor their gaits or behaviors. However, network unreliability and high power consumption have limited their applicability.In this work, we present an animal behavior recognition, classification and monitoring system based on a wireless sensor network and a smart collar device, provided with inertial sensors and an embedded multi-layer perceptron-based feed-forward neural network, to classify the different gaits or behaviors based on the collected information. In similar works, classification mechanisms are implemented in a server (or base station). The main novelty of this work is the full implementation of a reconfigurable neural network embedded into the animals collar, which allows a real-time behavior classification and enables its local storage in SD memory. Moreover, this approach reduces the amount of data transmitted to the base station (and its periodicity), achieving a significantly improving battery life. The system has been simulated and tested in a real scenario for three different horse gaits, using different heuristics and sensors to improve the accuracy of behavior recognition, achieving a maximum of 81%.

A Sensor Fusion Horse Gait Classification by a Spiking Neural Network on SpiNNaker

The study and monitoring of the behavior of wildlife has always been a subject of great interest. Although many systems can track animal positions using GPS systems, the behavior classification is not a common task. For this work, a multi-sensory wearable device has been designed and implemented to be used in the Donana National Park in order to control and monitor wild and semi-wild life animals. The data obtained with these sensors is processed using a Spiking Neural Network (SNN), with Address-Event-Representation (AER) coding, and it is classified between some fixed activity behaviors. This works presents the full infrastructure deployed in Donana to collect the data, the wearable device, the SNN implementation in SpiNNaker and the classification results.

A 20Mevps/32Mev event-based USB framework for neuromorphic systems debugging

Neuromorphic systems are engineering solutions that take inspiration from biological neural systems. They use spike-or event-based representation and codification of the information. This codification allows performing complex computations, filters, classifications and learning in a pseudo-simultaneous way. Small incremental processing is done per event, which shows useful results with very low latencies. Therefore, developing this kind of systems requires the use of specialized tools for debugging and testing those flows of events. This paper presents a set of logic implementations for FPGA that assists on the development of event-based systems and their debugging. Address-Event-Representation (AER) is a communication protocol for transferring events/spikes between bio-inspired chips/systems. Real-time monitoring and sequencing, logging and playing back long sequences of events/spikes to and from memory; and several merging and splitting ports are the main requirements when developing these systems. These functionalities and implementations are explained and tested in this work. The logic has been evaluated in an Opal-Kelly XEM6010 acting as a daughter board for the AER-Node platform. It has a peak rate of 20Mevps when logging and a total of 32Mev of logging capacity on DDR when debugging an AER system in the AER-Node or a set of them connected in daisy chain.

Accuracy Improvement of Neural Networks Through Self-Organizing-Maps over Training Datasets

Although it is not a novel topic, pattern recognition has become very popular and relevant in the last years. Different classification systems like neural networks, support vector machines or even complex statistical methods have been used for this purpose. Several works have used these systems to classify animal behavior, mainly in an offline way. Their main problem is usually the data pre-processing step, because the better input data are, the higher may be the accuracy of the classification system. In previous papers by the authors an embedded implementation of a neural network was deployed on a portable device that was placed on animals. This approach allows the classification to be done online and in real time. This is one of the aims of the research project MINERVA, which is focused on monitoring wildlife in Donana National Park using low power devices. Many difficulties were faced when pre-processing methods quality needed to be evaluated. In this work, a novel pre-processing evaluation system based on self-organizing maps (SOM) to measure the quality of the neural network training dataset is presented. The paper is focused on a three different horse gaits classification study. Preliminary results show that a better SOM output map matches with the embedded ANN classification hit improvement.

Live demonstration — Multilayer spiking neural network for audio samples classification using SpiNNaker

In this demonstration we present a spiking neural network architecture for audio samples classification using SpiNNaker. The network consists of different leaky integrate-and-fire neuron layers. The connections between them are trained using firing rate based algorithms. Tests use sets of pure tones with frequencies that range from 130.813 to 1396.91 Hz. Audio signals coming from the computer are converted to spikes using a Neuromorphic Auditory Sensor and, after that, this information is sent to the SpiNNaker board through a PCB that translates from AER to 2-of-7 protocol. The classification output obtained in the spiking neural network deployed on SpiNNaker is then shown in the computer screen. Different levels of random noise are added to the original audio signals in order to test the robustness of the classification system.

Comprehensive Evaluation of OpenCL-Based CNN Implementations for FPGAs

Deep learning has significantly advanced the state of the art in artificial intelligence, gaining wide popularity from both industry and academia. Special interest is around Convolutional Neural Networks (CNN), which take inspiration from the hierarchical structure of the visual cortex, to form deep layers of convolutional operations, along with fully connected classifiers. Hardware implementations of these deep CNN architectures are challenged with memory bottlenecks that require many convolution and fully-connected layers demanding large amount of communication for parallel computation. Multi-core CPU based solutions have demonstrated their inadequacy for this problem due to the memory wall and low parallelism. Many-core GPU architectures show superior performance but they consume high power and also have memory constraints due to inconsistencies between cache and main memory. OpenCL is commonly used to describe these architectures for their execution on GPGPUs or FPGAs. FPGA design solutions are also actively being explored, which allow implementing the memory hierarchy using embedded parallel BlockRAMs. This boosts the parallel use of shared memory elements between multiple processing units, avoiding data replicability and inconsistencies. This makes FPGAs potentially powerful solutions for real-time classification of CNNs. In this paper both Altera and Xilinx adopted OpenCL co-design frameworks for pseudo-automatic development solutions are evaluated. A comprehensive evaluation and comparison for a 5-layer deep CNN is presented. Hardware resources, temporal performance and the OpenCL architecture for CNNs are discussed. Xilinx demonstrates faster synthesis, better FPGA resource utilization and more compact boards. Altera provides multi-platforms tools, mature design community and better execution times.