A convolutional neural architecture: an application for defects detection in continuous manufacturing systems

One of the most important and critical aspects in all manufacturing processes is product inspection. Neural network based systems allow a compromise between resolution and processing speed in automatic inspection. This work introduces the development of a neural architecture, named Convolutional Top-Down Spiral Architecture, used to automatically generate digital filters for artificial vision inspection systems. Experimental results of this architecture applied for the detection of defects over paper pulp images gathered in a real environment are presented.