A pruned DSP96000 image classifier

The authors have developed a high-performance DSP96000 based image classification system that uses a pruned fast Fourier transform algorithm and a back-propagation classifier. The conventional image classification system extracts compact Fourier “spectral signatures” from images and applies them to back-propagation classifier. The pruned fast Fourier transform used in the proposed algorithm to generate the input to the backpropagation classifier is presented. The proposed system eliminates the generation of a digital sampling detector lookup table and the extraction of spectral feature vectors using the digital sampling detector. The complete hardware and software design is described. The DSP96002 Development System used to implement both the conventional and proposed image classifiers is developed. An interface circuit is designed to interface the DSP96002 processor on the DSP96002 Development Board to a frame grabber using DT-Connect. The DSP96002 assembler code for the proposed algorithm is described. It is shown that the computational time of the proposed algorithm is greatly reduced and the RMS error of the proposed algorithm is comparable with the conventional algorithm