Using Deep Attention Networks to Extract Defects in Crisscross Background | IEEE Conference Publication | IEEE Xplore