Neural vision sensors for surface defect detection

Vision sensors are built from a camera and intelligent hardware and/or software. Steadily decreasing microelectronic costs have spawned a large number of vision sensory applications, such as surface defect detection. A constructive method for defect detection entails a mixture of mathematical and intelligent modules. Such a heterogeneous modular system can be realized in many ways. In this paper we discuss a packet-switched implementation on a macro-enriched field-programmable gate-array.