A Shadow Detection Method Based on Neural Network in PCBA Height Measurement

In Fringe Projection Profilometry (FPP) systems, the shadow caused by the object which results in the occluded area without fringe pattern will lead to the reconstruction errors. In particular, in Printed Circuit Board Assembly(PCBA) height measurement applications, its complex surface characteristics and other phenomena such as secondary reflection of projected light increase the difficulty of shadow detection. In this paper, we propose a shadow detection method combining neural network model and image post-processing. This method obtains feature images containing shadow through a multi-directional fringe projection system. A neural network model is trained by extracting neighborhood features of shadow points, and then the output of the network is checked by a post-processing algorithms to complete shadow detection. The results of the experiments show that the Mean Intersection Over Union (MIoU) value of this method is 0.9205, and the Misclassification Error (ME) value is 0.0064, which are better than the state of art method and meets the actual inspection requirements.