Multispectral recognition techniques present status and a planned hybrid recognition system

Experiments with multispectral data have shown that it is possible to identify many desired terrain features. Such identifications often span large areas and are important in studies of crops, forestry, urban development, water resources, pollution control, hydrology, and many other aspects of ecological research. Experience has shown that high data processing speed is required for timely data reduction and application of the results. Techniques developed at The University of Michigan have solved the problem of keeping pace with the desired rate of recognizing objects in "real-time." However, three additional problems have been identified: 1. Setting up a processor such as the Michigan Processor (SPARC) requires considerable time. This time needs to be reduced. 2. Determination of an "optimum" setup cannot be done quickly or practically using such a processor by itself. 3. On line modification of the processing rule to account for natural variation seems necessary and requires computation which cannot be performed with an analog machine. A solution to these problems is proposed and discussed in this report. Specifications for a design approach are presented for a prototype hybrid multi-spectral processor (SPARC/H) having high throughput capability. The approach is to add an interacting digital machine to the system and reconfigure the analog machine design to be compatible with the digital system. The strength of our approach lies in the fact that algorithms have been developed which make effective use of the particular capabilitles of the two computers working together on a highly interactive basis. Study is still required to modify and develop these algorithms further for the digital machine and to determine the effectiveness of the man-machine interface. Study and specification of the final configuration needed for operational use can be pursued using this system as a tool.