Visual-based anomaly detection empowered by synthetic image generation
Abstract
Detecting anomalies in the manufacturing domain presents significant challenges due to the scarcity of defective samples. Generative models provide a promising solution by synthesizing realistic training data. However, most anomaly generation approaches rely exclusively on standard benchmark datasets and lack validation in real-world industrial settings. Given the wide variability in the type and appearance of anomalies in industrial settings, generated images should support downstream model training and also closely resemble actual defects. In this work, we propose a framework for anomaly generation and detection, using overhead cranes as a case study. We construct a custom dataset of overhead cranes collected from Aalto University Mechatronic Lab, including both normal and defective images, with various anomaly types such as surface abrasion, oil stains, and rust. Our approach revises the standard diffusion models by introducing attention-guided anomaly generation and the mask mechanism to enable the model to concentrate on generating the anomaly area. This method enables the generation of both realistic small-scale and large-scale anomalies with only a few training samples. Experimental results demonstrate that the generated anomalies closely resemble real-world crane defects and enhance the performance of downstream anomaly detection models. GitHub Repository: https://github.com/PengWuElliot/Thesis-Diffusion.
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.