Hardware implementation of a neural network performing multispectral image fusion

摘要

A prototype system has been developed to demonstrate the use of neural network hardware to fuse multispectral imagery. Research in support of this consists of neural network simulations fusing 4 to 7 bands of Landsat imagery for terrain classification and other simulations fusing multiple bands of synthetic imagery for subpixel target detection and classification. The authors have developed a system hosted by a PC, and consisting of a circuit board containing a neural network chip and all associated circuitry, 3 input/output boards, and software to demonstrate the fusion of multispectral imagery. The neural network chip and board have been designed and fabricated, and are currently being integrated with the other hardware components and the software. Section 1 describes the problem domain. Section 2 describes some software simulations in support of this effort and the test results utilizing Landsat imagery and synthetic subpixel target imagery. Section 3 describes the architecture of the system being built to demonstrate the use of neural network hardware to process multispectral imagery. Section 4 summarizes results to date and plans for the future.