Inspection of flexible printed circuit board (FPCB) is a crucial process in the manufacture industry. Besides, edge detection and extraction of the region of interest (ROI) is always a main issue. An effective segmentation method helps to reduce the influence of the irrelevant factors, thus largely augment the performance of defect inspection in the following process. In this paper, a method of segmentation of the gold finger area of FPCB was proposed to solve the problem of adhesion between gold finger and cover film. First, a Gaussian filter was applied to the grayscale image. Second, a blurry edge image was obtained by the residual image between the Gaussian image and the other after an erosion operation. Third, a binary image was created by using an adaptive threshold method, and finally, with the application of histogram statistic, the rectangle area of gold finger was extracted. Experimental results show that the proposed method accurately locate the gold finger area for our FPCB images. In addition, it takes 2.654s to extract the ROI with the adaptive threshold and 1.442s without it. By feeding it to a classifier, a total time of 3.903s was spent to analyze the possible defects, which means the algorithm is suitable for automated optical inspection (AOI) for FPCBs.

In multi-task reinforcement learning there are two main challenges: at training time, the ability to learn different policies with a single model; at test time, inferring which of those policies applying without an external signal. In the case of continual reinforcement learning a third challenge arises: learning tasks sequentially without forgetting the previous ones. In this paper, we tackle these challenges by proposing DisCoRL, an approach combining state representation learning and policy distillation. We experiment on a sequence of three simulated 2D navigation tasks with a 3 wheel omni-directional robot. Moreover, we tested our approach's robustness by transferring the final policy into a real life setting. The policy can solve all tasks and automatically infer which one to run.