Machine learning technology applied to production lines: Image recognition system

The recent trend toward mass customization has increased the demand for multiproduct/multivolume production and driven a need for autonomous production systems that can respond quickly to changes on production lines. Production facilities using cameras and robot-based image recognition technologies must also be adaptable to changes in the image-capturing environment and product lots, so technology enabling the prompt generation and well-timed revision of image-processing programs is needed. The development of image recognition systems using machine learning techniques has been progressing with the aim of constructing such autonomous production systems. Furthermore, in addition to the need for automatic generation of image-processing programs, the development of technology for automatically and quickly detecting changes in the production environment to achieve a stable production line has also become an issue. We have developed technology for generating preprocessing programs, extracting image feature values, and optimizing learning parameters and have applied this technology to template matching widely used in image processing and to product accept/reject testing. We have also developed technology for sensing changes in the image-capturing environment by using images captured at the time of learning as reference and detecting changes in subsequent image feature values. These technologies enable the generation of various types of image-processing programs in a short period of time and the detection of signs of change in the image-capturing environment before the recognition rate drops.