Multi-Scale Defect Detection Using Modified Faster R-CNN for Plates with Complex Surfaces

Defect recognition plays a crucial role in investigating a panel. Mostly the recent manual investigating methods are time consuming and high expense. The fast recurrent neural network is enhanced and is less time consuming. The faster R-CNN has been proposed in the research work. A feature pyramid network associated with ResNet-50 has been efficiently able to detect the defects in a precise manner. In this manuscript for localization of defects we have used Region of Interest Align in place of pooling ROI. Then an enhanced feature network has been used to precisely detect the defects. Therefore, the k-means clustering algorithm is used to cluster the defects so that the defects can be easily detected. In this paper the data set has been taken and the algorithms are compared with the existing algorithm to check the accuracy and efficiency of the proposed system. The detection accuracy has been converted and detected properly and validated in the paper.