| 1. |
- Malki, Suleyman, et al.
(författare)
-
A CNN hardware-only approach to speed measurement
- 2006
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Vision sensors are of increasing interest for non-destructive, remote measurements in industrial as well as consumer products. Their operation is based on intelligent techniques for image understanding, followed by robust quantification of the detected phenomenon. A well-known application area is movement detection and isolation. This paper discusses how movement can be detected and quantified by means of a single Cellular Neural Network, implemented as a Network-on-Chip and realized on a FPGA. The final hardware can measure up a moving object at a speed of 250 frames per second.
|
|
| 2. |
- Malki, Suleyman, et al.
(författare)
-
A CNN-Specific Integrated Processor
- 2009
-
Ingår i: Eurasip Journal on Advances in Signal Processing. - : Springer Science and Business Media LLC. - 1687-6172 .- 1687-6180.
-
Tidskriftsartikel (refereegranskat)abstract
- Integrated Processors (IP) are meant to supply algorithm-specific cores to a micro-electronic system. They are usually developed separately, and upon success turned into a part, that either by programming or by configuration can be re-used within many systems. This paper looks at architectures for Cellular Neural Networks (CNN) to become realized as IP. First current digital implementations are reviewed, and the memory-processor bandwidth issues are analyzed. Then a generic view is taken on the structure of the network, and a new intra-communication protocol based on rotating wheels is proposed. It is shown that this provides for guaranteed high-performance with a minimal network interfaces. The resulting node is small and supports multi-level CNN designs, giving the system a 30-fold increase in capacity compared to classical designs. As it facilitates multiple operations on a single image, and single operations on multiple images, with minimal access to the external image memory, balancing the internal and external data transfer requirements can optimize the system operation. Special consideration is given to the treatment of the boundary nodes. In conventional digital CNN designs, such requires additional logic to handle the CNN value propagation scheme. In the new architecture, only a slight modification of the existing cells is necessary to model the boundary effect. A typical prototype for visual pattern recognition will house 4096 CNN cells with a 2% overhead for making it an IP.
|
|
| 3. |
- Malki, Suleyman, et al.
(författare)
-
A DT-CNN Data-Flow Implementation
- 2008
-
Ingår i: 2008 11TH INTERNATIONAL WORKSHOP ON CELLULAR NEURAL NETWORKS AND THEIR APPLICATIONS. ; , s. 17-22
-
Konferensbidrag (refereegranskat)abstract
- Digital implementations of Cellular Neural Networks are studied in terms of their communication requirements. Secure and reliable communication seems to imply close control, which degrades performance. We introduce a mechanism that removes the need for explicit local network control, taking the internal network communication out of the performance equation. This allows handling boundary conditions without introducing additional cells and facilitates multi-level implementations. A typical feature extraction task in hand vein recognition shows a 20x performance improvement for the Cellular Neural Network implementation.
|
|
| 4. |
- Malki, Suleyman, et al.
(författare)
-
An exploration of digital CNN implementations
- 2004
-
Ingår i: Proceedings SSOCC.
-
Konferensbidrag (refereegranskat)abstract
- Computationally hard problems, such as operations on n-dimensional maps, are in need of efficient solutions. Cellular Neural Networks have this promise. This paper explores digital realizations of such computational paradigms through the case of real-time image processing. It is shown that a behaviour-based spatial unrolling of the time-critical numerical convergence loop outperforms a classical function-based architecture. An FPGA implementation on Xilinx Virtex-II illustrates a 50 times higher throughput, bringing intelligent vision into reach.
|
|
| 5. |
- Malki, Suleyman, et al.
(författare)
-
Bit-Serial Approach Moves DT-CNN into Real-Time
- 2007
-
Ingår i: Proceedings SSoCC2007.
-
Konferensbidrag (refereegranskat)abstract
- The Cellular Neural Network provides an interesting template for non-linear signal processing. It is essentially based on 2nd-order (reaction / diffusion) differential equations. Harrer and Nossek have introduced the discrete-time equivalent that paves the ground for digital implementation. The interconnect requirements seem to have made it impossible to realistically implement this model. More than emulation is not possible for a network based on word-serial communication. In this paper, it is discussed that a bit-serial communication scheme brings a more realistic implementation. Furthermore, using also bit-serial techniques within each node permits to decrease the footprint and therefore makes a 2-level network for the direct implementation of reaction / diffusion mechanism possible.
|
|
| 6. |
- Malki, Suleyman, et al.
(författare)
-
CNN image processing on a Xilinx Virtex-II 6000
- 2003
-
Ingår i: Proceedings of the 16th European Conference on Circuit Theory and Design, ECCTD'03. - 8388309951 ; , s. 4-261
-
Konferensbidrag (refereegranskat)abstract
- Image processing is one of the popular applications of Cellular Neural Networks. Macro enriched field-programmable gate-arrays can be used to realize such systems on silicon. The paper discusses a pipelined implementation that supports the handling of gray-level images at 180 to 240 Mpixels per second by exploiting the Virtex-II macros to spatially unroll the local feedback
|
|
| 7. |
- Malki, Suleyman, et al.
(författare)
-
Design Space Exploration for a DT-CNN
- 2008
-
Ingår i: 2008 11TH INTERNATIONAL WORKSHOP ON CELLULAR NEURAL NETWORKS AND THEIR APPLICATIONS. ; , s. 69-74
-
Konferensbidrag (refereegranskat)abstract
- The paper presents design considerations for a digital Cellular Neural Network. The architectural spectrum is categorized from a rich variety in nodal structural transformations and formulated by an exploration model. This model is then applied to study the influence of the memory band-gap. Experiments show the benefits of a tiled implementation, mixing spatial network parallelism with time-multiplexed nodes based on a local register stack.
|
|
| 8. |
- Malki, Suleyman, et al.
(författare)
-
Design Space Exploration for the integrated digital CNN Camera
- 2008
-
Ingår i: 1st International Conference on Information Technology, 2008. - 9781424422449 ; , s. 107-110
-
Konferensbidrag (refereegranskat)abstract
- The main thrust of CNN technology has been in analog implementations. Early attempts for digital implementations have a limited potential, confining the usage to for instance linear algorithms for B/W pictures. The paper looks into the prospects of successfully integrating the full CNN potential. This requires a carefully tuning of memory and network-on-chip bandwidth to the temporal/spatial aspects of the neural nodes. The paper introduces an exploration model and discusses architecture with application to streaming grey-scale images.
|
|
| 9. |
|
|
| 10. |
- Malki, Suleyman
(författare)
-
Discrete-Time Cellular Neural Networks Implemented on Field-Programmable Gate-Arrays to Build a Virtual Sensor System
- 2006
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Image processing is one of the popular applications of Cellular Neural Networks. Macro enriched field-programmable gate-arrays can be used to realize such systems on silicon. At first glance a pipelined approach, based on circuit switching, seems promising. The digital implementation supports the handling of grey-level images at 180 to 240 Mpixels per second by exploiting the Xilinx Virtex-II macros to spatially unroll the local feedback. Later on, in order to overcome design limitations and thus enhance performance, the benefits of packet switching have been explored. The digital implementation is performed using Xilinx Virtex-II Pro P30. The advantages over the approach of circuit switching are discussed. Finally, the thesis illustrates the power of the different implementations experimentally. It is shown how these implementations can be used to measure from images or to create dynamic, autonomous processes that facilitate measurements within topographic maps. Applications range from image understanding to robot navigation.
|
|
