Container Inspection AI Training From Fault Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing methods for inspecting containers in drinks processing systems are time-consuming and costly due to the need for expert setup and extensive training of neural networks with marked images.
Innovation Solution
A method that utilizes a conventional image processing method to classify camera images, compile fault images and fault-free images, and train a second evaluation unit with artificial intelligence on site using a specific training data set, reducing the need for expert intervention and time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional image processing methods are used to inspect containers, then fault detection is achieved, but expert setup time and cost increase
Solution Approach 1:
The system automatically generates training data using the conventional image processing method itself, eliminating the need for external expert annotation. The AI model trains on self-generated data, making the system self-sufficient and removing the time-consuming expert setup phase while maintaining detection reliability
Solution Approach 2:
The system performs preliminary automated data collection and training data generation before AI model deployment. By pre-generating training datasets through automated container inspection and fault classification, the system prepares everything needed for AI training in advance, eliminating subsequent expert intervention requirements
2Measurement precision
If neural networks are trained with extensively marked images, then detection accuracy improves, but training time and cost increase
Solution Approach 1:
The system automatically generates training datasets by using the conventional image processing method to inspect containers, classify faults, and create marked training images without human intervention. This self-service approach maintains high detection accuracy while dramatically improving training efficiency by eliminating manual marking processes
Solution Approach 2:
The conventional image processing method acts as an intermediary that automatically creates quality training data for the AI model. Instead of requiring experts to mark images, the system uses the established conventional method as a mediator to generate accurate training datasets, bridging the gap between conventional inspection and AI-based inspection
Data Source
AI summary
A method for optically inspecting containers in a drinks processing system, wherein the containers are transported as a container mass flow using a transporter and captured as camera images by an inspection unit arranged in the drinks processing system, and wherein the camera images are inspected for faults by a first evaluation unit using a conventional image processing method, wherein the camera images with faulty containers are classified as fault images and the faults are correspondingly assigned to the fault images as fault markings, wherein the camera images with containers considered to be good quality are classified as fault-free images, the fault images, the fault markings and the fault-free images are compiled as a specific training data set, and wherein, using the specific training data set, a second evaluation unit is trained in situ with an image processing method working on the basis of artificial intelligence.