Edge Detection in Dark Industrial Environments

To solve the problem that current edge detectors cannot accurately detect the edges of artifacts under low-light conditions, we propose an edge detection network-EDD that adapts to scale changes and low-light environments. First, we utilize a feature pyramid to divide the features of the input image into 4 stages of different resolutions. Second, we introduce a scale enhancement module to obtain feature information at different scales. In addition, in order to solve the challenge of inconspicuous edge details in low-light environments, we design a high frequency enhancement module. We conduct experimental verification on the BSDS500 dataset and our own artifact dataset MPBD, and the results show that EDD has a significant edge detection effect in low-light environments. Compared with other methods, EDD can capture the edge of the workpiece more accurately, and provides an effective solution to the edge detection problem in dark conditions.