Pattern recognition scanning for plant inspection - rotating cylindrical objects about longitudinal axis under scanning camera to generate image for input to evaluation circuit, and cross-correlation used to train neural network

摘要

The pattern recognition method involves rotating cylindrical objects about their longitudinal axis under a scanning camera (16) producing a linear scan which is input to evaluation circuit (17). A small sample (18) is processed in a cross-correlator (19) with a comparison sample (20) for which the data is passed through the system to achieve a maximum correlation. This sets a reference point for processing by a trained neural network. The cross-correlator uses a Hough transform, a special case of the FFT. The pattern recognition is principally by means of corner or edge recognition detection and the recognition requires the determination of the radii of curvature. USE/ADVANTAGE - for automatic inspection systems, quality control systems partic. determining need for more herbicide to kill off weeds among plants. Flexible system adaptable for several applications.