Anomaly Detection For Industrial Application: Spotlight On Patchcore And Autoencoder Insights With Resnet Backbone

Abstract

Ensuring product quality is one of the key requirements in industrial manufacturing, where early detection of defects helps in reducing losses, improving safety, and maintaining customer satisfaction. Manual inspection is often time-consuming and prone to human error, which has motivated the use of Artificial Intelligence (AI) for automated anomaly detection. This project focuses on developing a basic yet effective anomaly detection system using image data. Two simple deep learning approaches are employed: an Autoencoder and a Patch-based analysis method.

This project focuses on building a simple anomaly detection system for industrial applications. The goal is to identify whether a product image is normal or defective. To achieve this, we use a basic Convolutional Neural Network (CNN) for feature extraction and apply two easy approaches.

A small dataset of normal and defective product images is used for testing. The project demonstrates that even with basic deep learning techniques, it is possible to detect defects in industrial images with good accuracy. This highlights the usefulness of AI in improving product quality and reducing manual inspection effort.