Implementation of FPGA-based artificial neural network for character recognition
Artificial Neural Networks (ANN) are non-linear applied math knowledge data modeling tools, usually used model advanced relationships between inputs and outputs or to seek out patterns in data. A generic hardware primarily based ANN is planned and executed using VHDL coding. This project may be see...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Subjects: | |
| Online Access: | http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/59425/1/Page%201-24.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/59425/2/Full%20text.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Artificial Neural Networks (ANN) are non-linear applied math knowledge data modeling
tools, usually used model advanced relationships between inputs and outputs or to seek out patterns in data. A generic hardware primarily based ANN is planned and executed using VHDL coding. This project may be seen as a place to begin for learning ANN. It explores in approach a hardware-based application of ANN employs FPGA. The sixteen
toggle switches are given as input while the end product is exhibited on the LCD display.
This classifier is trained to identify letters on a 4x4 binary grid filled by a user through 16 toggle switches. The most probable class suggested by the ANN is displayed on an LCD screen. To demonstrate the practicality of FPGA execute of ANN, the ANN trained to
acknowledge twenty English and nine Arabic character patterns on a 4x4 grid. In structural of ANN, the used of three-layer is implemented entirely with 32-bit single exactitude floating purpose arithmetic to ensure flexibility and accuracy for its wide selection of applications. The resulting design file is programmed into the Altera Cyclone II FPGA on the Altera DE2 development and education board. The design also includes a training supervisor that trains the ANN recognized the total of 29 English and Arabic alphabet predefined characters. The result is promising as the ANN is able to recognize all
characters defined to training characters patterns. Each alphabet is tested in 20 English
alphabet and 9 Arabic alphabet, after implementation, are done and performance issues of
the design are analyzed. The output gives good results, and finally this project shows the
flexibility and also the endless chance of hardware primarily based implementation of
ANN , the achievement of recognition rate for alphabet English and Arabic character are
76.92% and 32.14% respectively. |
|---|