Microelectronic industry is the kernel and foundation of the information industry, it is the important assurance to improve national economy. With the development of microelectronic products(integrated circuit, printed circuit board, etc) directing to high density, thin separation and low defect ratio, its inspection requirement is higher on aspects of precision, efficiency, universal, and intelligence etc. Therefore, this paper researched on the general key techniques in the field of microelectronic products vision inspection, covered the shortage of traditional inspection on aspects of fast and precision locating, image mosaic, and fine defect test, completed theory study on physical dimension and defect inspection of microelectronic products based on machine vision, developed the prototype and used lots of experiments to prove its correctness and feasibility. The main research works in this dissertation are as follows:(1) Based on the automatic vision inspection principle and modern inspection requirement, designed a new vision inspection system for microelectronic products (PVMS). According to the features of precision, efficiency, inspection dimension and accuracy varying great range, designed closed-loop control system composition of servomechanism and grating to assure system's precision, smooth, efficiency, and to realize the great dimension range measurement; designed image capture system composition of CCD, automatic variable power lens and self-adaptive lighting, to realize integration of large inspection precision range; designed electric control and software system to make the inspection convenience and fast. Finally, developed the prototype instrument.(2) Considering the problem that genetic algorithm running result is affected greatly by the initial parameter and it's easy to trap in local optimum, based on its mathematic theory, studied deeply on self-adaptive genetic algorithm, presented a new viewpoint -- superiority inheritance, furthermore, designed adaptive genetic algorithm based on superiority inheritance. The experiment showed that this algorithm could find the global best solution, and had strongly global hunting capability. It is prior to traditional adaptive genetic algorithms on aspects of accuracy, stability, and repeatability etc. It was applied in fast image