 This paper proposes a new approach for automatically detecting metallic defects in industrial settings. It uses a cascade autoencoder, KC, to segment and classify defects in the input image. The KC transforms the input image into a pixel-wise prediction mask, which is then used by a compact convolutional neural network, CNN, to classify the defects. The proposed method was tested on an industrial dataset and demonstrated its effectiveness in detecting metallic defects, under different conditions. This article was authored by Xian Tao, Da Peng Zhang, Wen Ji Ma, and others.