DefectNet: Toward Fast and Effective Defect Detection
Li Feng, Feng Li, Qinggang Xi
Abstract
The existing object detection algorithms based on the convolutional neural network (CNN) are always devoted to the detection of natural objects and have achieved admirable detection effects. At present, these detection algorithms have been applied to the detection of defect data. In fact, the detection of defect data is different from the detection of general natural object data. First, it has a large number of images without annotations (that is, normal images), and they each contain different background information. Second, its processing principles are fundamentally different from general object detection problems. Therefore, the application of a general object detection algorithm based on CNN may not be perfect in this problem. In this article, a novel defect detection network (DefectNet) is proposed to solve the problem of defect detection. It first uses a shared weight binary classification network to determine whether an image contains the targets and then uses the detection network to detect the targets. Theoretical deduction and experimental results fully confirm that it can effectively improve the detection speed and effect of the general object detection network based on CNN. (Our code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/li-phone/DefectNet.git</uri> )