Abstract:The semantic segmentation network based on U-Net (U-shaped network structure) in Si 3 N 4 ceramic bearing cylindrical roller surface defect segmentation has the problems that low accuracy of defect feature extraction and large number of model parameters. It cannot meet the segmentation accuracy requirements of different size defects. The semantic segmentation network (D-A-IU-Net-Semantic segmentation network model with embedded multi-scale defective features) with embedded multi-scale defective feature fusion module is proposed. D-A-IU-Net inherits the encoding–decoding features of U-Net semantic segmentation network in structure and has the ability to sense multi-layer defective features. To reduce the number of model parameters and the network model training time, the D-A-IU-Net network model is downsampled only three times. To address the loss of detailed semantic information during downsampling, the network adopts a depthwise separable convolution layer with stride = 2 instead of the maximum pooling layer in the coding module. The segmentation accuracy is further improved. For overcoming the low segmentation accuracy of tiny defects on Si 3 N 4 ceramic bearing cylindrical roller surface, the Atrous Spatial Pyramid Pooling Module (ASSP module-Hollow space convolution pool pyramid) is added to the bottom of the D-A-IU-Net network model to replace the model structure at the bottom of the U-net network model. To avoid the performance degradation caused by the increasing depth of the network model, two identity blocks are added at the bottom. A dataset of Si 3 N 4 ceramic bearing cylindrical roller surface defects is prepared for validation. The experimental results show that the mIoU (mean Intersection over Union) of the D-A-IU-Net network model reaches 90.70%, the mPA (mean Pixel Accuracy) is improved to 95.13%, and the single image prediction time is 46.53 ms. The region segmentation efficiency of Si 3 N 4 ceramic bearing cylindrical roller surface defects is improved.

A spectrum-domain instance segmentation model for casting defects

Segmentation of Casting Defect Regions for the Extraction of Microstructural Properties.

Interactive Defect Segmentation in X-Ray Images Based on Deep Learning

Casting defect region segmentation method based on dual-channel encoding–fusion decoding network

Real-Time Forecasting of Subsurface Inclusion Defects for Continuous Casting Slabs: A Data-Driven Comparative Study

Automated Detection of Defects with Low Semantic Information in X-ray Images Based on Deep Learning

A sparse-representation-based robust inspection system for hidden defects classification in casting components.

Progressive Frequency-Guided Depth Model with Adaptive Preprocessing for Casting Defect Detection

Attention-based Stacked Supervised Poisson Autoencoders for Defects Prediction in Casting-rolling Process

A Sub-Region Unet for Weak Defects Segmentation with Global Information and Mask-Aware Loss.

Latent Space Segmentation Model for Visual Surface Defect Inspection

An Adaptive Image Segmentation Network for Surface Defect Detection

An Instance Segmentation Method for Insulator Defects Based on an Attention Mechanism and Feature Fusion Network

A Casting Surface Dataset and Benchmark for Subtle and Confusable Defect Detection in Complex Contexts

AnomalySeg: Deep Learning-Based Fast Anomaly Segmentation Approach for Surface Defect Detection

Nondestructive testing method of Si3N4 ceramic bearing cylindrical roller surface defects based on depthwise separable convolution coupled with identity block ASPP

Detection of a casting defect tracked by deep convolution neural network

LERENet: Eliminating Intra-class Differences for Metal Surface Defect Few-shot Semantic Segmentation

Performance, Successes and Limitations of Deep Learning Semantic Segmentation of Multiple Defects in Transmission Electron Micrographs

A nondestructive automatic defect detection method with pixelwise segmentation

Improving Scattered Defect Grading in Castings Digital Radiographs via Smoothing the One-Hot Encoding