Hybrid CNN-LSTM Architecture for Automated Defect Detection in Industrial Surface Inspection

Automating defect detection in the industrial surface inspection is significant for a product quality and efficiency of the manufacturing process. Inspection techniques that used the conventional methodologies take a long time to complete and are less accurate; hence, the need to use deep learning solutions. This paper presents Hybrid CNN-LSTM Architecture which combined the CNN for spatial feature extraction and LSTM for temporal feature recognition. The employment of CNN and LSTM together improves the recognition capacity of the spatial and temporal features for better defect categories and segmentation. The experiments show that the proposed method improves the defect detection accuracy by 20% and the false positive rates by 30% than the existing method. The framework performs well in detecting the defects on various surface texture type and materials, which makes the framework to be a good solution for the industrial quality control.