A Research Survey on Intelligent Detection Surface Irregularities with Deep Learning

Steel part surface detection is a task in multiple industries, such as agriculture, manufacturing, construction, aerospace, mining, etc. Intelligent Detection of Surface Irregularities includes visual recognition of various steel surface defects on steel parts that are present at inaccessible locations. This study works on the recognition of two major irregularities, which are corrosion and cracking. Corrosion and cracking of metal components cost around 4% of GDP for every nation each year, making it a major problem. Corrosion and cracks must be detected within time to perform maintenance, which assures the reliability of the structure. This study includes solutions to direct irregularity detection method that works on the visual aspects of irregularities. Deep learning methods are used to extract attributes and features from images. Our approach to classification based on irregular and normal surfaces is initiated by collecting various datasets of images containing irregularities of steel surfaces. Pre- processing using OpenCV includes things such as resizing images to the normal standard size, normalization, and edge detection. Feature extraction using sci-kit-image, OpenCV, CNN models with image segmentation techniques are trained to segment regions of an image with specific colors. TensorFlow is used to build and train convolutional neural networks (CNNs) and pretrained models such as VGG and Resnet are used for transfer learning. Performance is tested using metrics such as precision, recall, F1, and accuracy. Results show that the proposed system achieves a high level of accuracy and precision.