Anomaly Detection Threshold Calibration via Synthetic Image Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting anomalies in digital images of products require a time-consuming learning process to determine threshold values, necessitating a large number of good product images and struggling to predict the required number, which leads to inefficiencies in production time and false reject rates.
Innovation Solution
A method that generates digital images of similar products with known anomalies, subdivides images into regions, and uses statistical estimation to determine detection rates, allowing for the quick and efficient specification of threshold values to balance false reject and detection rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of good product images are collected for the learning process to determine threshold values, then the reliability of anomaly detection is improved, but the loss of production time increases
Solution Approach 1:
The patent applies preliminary action by performing the threshold value determination and detection rate estimation in advance through a test process with artificially created bad images. The learning process uses a small number of good images combined with synthetically generated bad images to pre-determine threshold values and detection rates before actual production inspection begins. This allows the system to be prepared and calibrated beforehand, avoiding time-consuming data collection during production runs.
2Productivity
If the number of good product images used in the learning process is reduced to save production time, then the productivity is improved, but the measurement precision of detection rate determination deteriorates
Solution Approach 1:
The patent applies copying by creating artificial copies of bad products through digital image manipulation. Real bad product images are processed to generate multiple variant images with the same anomalies, effectively multiplying the training data without requiring additional physical products. This allows the system to achieve sufficient statistical precision for detection rate determination while using a limited number of actual product images.
Solution Approach 2:
The patent applies parameter changes by systematically varying parameters in the artificial image generation process, such as anomaly position, size, and image characteristics. By changing these parameters across multiple generated images, the system accumulates diverse training data that improves measurement precision without requiring a proportional increase in the number of physical product images.
3Measurement precision
If threshold values are determined through a comprehensive learning process with many images, then the detection precision is improved, but the device complexity increases
Solution Approach 1:
The patent applies taking out by separating the threshold determination process into a distinct test phase that is extracted from the normal production inspection flow. The complex learning and threshold calibration operations are performed once in advance using a small dataset, and the resulting threshold values are then reused for all subsequent inspections. This extraction eliminates the need for continuous complex processing during production.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables simple and low-effort determination of detection rates for anomalies, optimizing the balance between false reject and detection rates, thereby improving production efficiency and safety.
Implementation Method 1
inspection devices are used in practice which irradiate the product with electromagnetic radiation, in particular with radiation in the x-ray spectrum
Implementation Method 2
the radiation penetrates deep enough into the product being examined and is 'reflected' in the product, wherein this 'reflection' is physically caused by a dispersion of the radiation penetrating in a volume region
Implementation Method 3
the term 'gray scale value' is used for the information that the detector generates as a function of the radiation power incident on the individual pixels or the corresponding radiation energy that is detected during the relevant exposure time
Data Source
AI summary
The invention relates to a method for detecting anomalies in digital images of products. A digital image is divided into regions, wherein a region is detected as a maximum anomaly if the value of the at least one property is greater than a predetermined maximum threshold value, or as a minimum anomaly if the value of the at least one property is less than a predetermined minimum threshold. In a test process, a plurality of digital bad images is generated from real or fictitious bad products, each which of which has at least one already known anomaly. Every bad image is divided into regions and the maximum value of the relevant property of the regions is determined as the maximum sample value of a maximum value sample or a minimum sample value of a minimum value sample. A detection rate for the at least one already known anomaly is determined from a sample generated in this way. For this purpose, according to one alternative, a suitable predetermined probability density function for describing the maximum value sample or the minimum value sample is parameterized using a statistical estimation method. The detection rate can thereby be calculated by integrating the parametrized probability density function using the predetermined maximum threshold value or the minimum threshold value as an integration limit. According to another alternative, the detection rate can be determined as a ratio of the number of values of the maximum value sample or the minimum value sample, which are greater than or equal to the predetermined maximum threshold value or less than or equal to predetermined minimum threshold value, and the total number of values of the maximum value sample or the minimum value sample. The detection rate that is detected in this manner can then be assigned to the at least one already known anomaly.