Constructing Efficient Features for Back Propagation in Visual Inspections

Back propagation(BP) is one of the most widely used artificial neural networks(ANNs) because of its powerful problem solving capability and simplicity. However, one of the inherent drawbacks of the BP algorithm is its slow convergence speed. When applied to visual inspections in machine vision, the large amount of the image data may make the convergence of the BP algorithm unpredictable.