Hyperspectral Imaging (HI) assembles high resolution spectral information from hundreds of narrow bands across the electromagnetic spectrum, thus generating 3D data cubes in which each pixel gathers the spectral information of the reflectance of every spatial pixel. As a result, each image is composed of large volumes of data, which turns its processing into a challenge, as performance requirements have been continuously tightened. For instance, new HI applications demand real-time responses. Hence, parallel processing becomes a necessity to achieve this requirement, so the intrinsic parallelism of the algorithms must be exploited. In this paper, a spatial-spectral classification approach has been implemented using a dataflow language known as RVCCAL. This language represents a system as a set of functional units, and its main advantage is that it simplifies the parallelization process by mapping the different blocks over different processing units. The spatial-spectral classification approach aims at refining the classification results previously obtained by using a K-Nearest Neighbors (KNN) filtering process, in which both the pixel spectral value and the spatial coordinates are considered. To do so, KNN needs two inputs: a one-band representation of the hyperspectral image and the classification results provided by a pixel-wise classifier. Thus, spatial-spectral classification algorithm is divided into three different stages: a Principal Component Analysis (PCA) algorithm for computing the one-band representation of the image, a Support Vector Machine (SVM) classifier, and the KNN-based filtering algorithm. The parallelization of these algorithms shows promising results in terms of computational time, as the mapping of them over different cores presents a speedup of 2.69x when using 3 cores. Consequently, experimental results demonstrate that real-time processing of hyperspectral images is achievable.
A generic bio-inspired adaptive architecture for image compression suitable to be implemented in embedded
systems is presented. The architecture allows the system to be tuned during its calibration phase. An evolutionary
algorithm is responsible of making the system evolve towards the required performance. A prototype has been
implemented in a Xilinx Virtex-5 FPGA featuring an adaptive wavelet transform core directed at improving
image compression for specific types of images.
An Evolution Strategy has been chosen as the search algorithm and its typical genetic operators adapted to
allow for a hardware friendly implementation. HW/SW partitioning issues are also considered after a high level
description of the algorithm is profiled which validates the proposed resource allocation in the device fabric.
To check the robustness of the system and its adaptation capabilities, different types of images have been
selected as validation patterns. A direct application of such a system is its deployment in an unknown environment
during design time, letting the calibration phase adjust the system parameters so that it performs efcient image
compression. Also, this prototype implementation may serve as an accelerator for the automatic design of
evolved transform coefficients which are later on synthesized and implemented in a non-adaptive system in the
final implementation device, whether it is a HW or SW based computing device.
The architecture has been built in a modular way so that it can be easily extended to adapt other types of
image processing cores. Details on this pluggable component point of view are also given in the paper.
KEYWORDS: Object recognition, Field programmable gate arrays, Neural networks, Computer architecture, Very large scale integration, Current controlled current source, Pattern recognition, Network security, Control systems, Network architectures
Neural networks, widely used in pattern recognition, security applications and robot control have been chosen for the task of object recognition within this system. One of the main drawbacks of the implementation of traditional neural networks in reconfigurable hardware is the huge resource consuming demand. This is due not only to their intrinsic parallelism, but also to the traditional big networks designed. However, modern FPGA architectures are perfectly suited for this kind of massive parallel computational needs. Therefore, our proposal is the implementation of Tiny Neural Networks, TNN -self-coined term-, in reconfigurable architectures. One of most important features of TNNs is their learning ability. Therefore, what we show here is the attempt to rise the autonomy features of the system, triggering a new learning phase, at run-time, when necessary. In this way, autonomous adaptation of the system is achieved. The system performs shape identification by the interpretation of object singularities. This is achieved by interconnecting several specialized TNN that work cooperatively. In order to validate the research, the system has been implemented and configured as a perceptron-like TNN with backpropagation learning and applied to the recognition of shapes. Simulation results show that this architecture has significant performance benefits.
In this paper we propose an architecture for an embedded traffic signal detection system. Development of Advanced Driver Assistance Systems (ADAS) is one of the major trends of research in automotion nowadays. Examples of past and ongoing projects in the field are CHAMELEON ("Pre-Crash Application all around the vehicle" IST 1999-10108), PREVENT (Preventive and Active Safety Applications, FP6-507075, http://www.prevent-ip.org/) and AVRT in the US (Advanced Vision-Radar Threat Detection (AVRT): A Pre-Crash Detection and Active Safety System). It can be observed a major interest in systems for real-time analysis of complex driving scenarios, evaluating risk and anticipating collisions.
The system will use a low cost CCD camera on the dashboard facing the road. The images will be processed by an Altera Cyclone family FPGA. The board does median and Sobel filtering of the incoming frames at PAL rate, and analyzes them for several categories of signals. The result is conveyed to the driver.
The scarce resources provided by the hardware require an architecture developed for optimal use. The system will use a combination of neural networks and an adapted blackboard architecture. Several neural networks will be used in sequence for image analysis, by reconfiguring a single, generic hardware neural network in the FPGA. This generic network is optimized for speed, in order to admit several executions within the frame rate. The sequence will follow the execution cycle of the blackboard architecture.
The global, blackboard architecture being developed and the hardware architecture for the generic, reconfigurable FPGA perceptron will be explained in this paper. The project is still at an early stage. However, some hardware implementation results are already available and will be offered in the paper.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.