Do you have any questions?

Classic image processing often works with fixed, predefined rules, for example to detect edges, contrasts, colours or geometric features. These methods are particularly suitable for clearly describable inspection tasks under controlled conditions.

AI-based image processing (vision AI), by contrast, learns relevant features from training images. This allows it to recognise complex visual relationships that are hard to describe with fixed rules. This is a particular advantage with varying products, surfaces, lighting conditions or defect patterns.

DENKweit develops its own vision AI technologies for this purpose, which can be trained with images from real production environments.

YOLO is a widely used family of AI models for object detection. It can be a suitable technical basis for many applications. However, a single model does not cover the entire life cycle of an industrial image processing solution.

Using our technologies has several advantages:

1. You can adapt and retrain a DENKnet to your challenges and requirements at any time.

2. We are at your side for all questions (AI training, integration, etc.), so that your AI deployment really runs the way you need it to.

3. With a DENKnet, you have a reliable and stable AI on an ongoing basis.

The advantage therefore lies not only in the AI model, but in the combination of in-house technology, ease of use and long-term industrial usability.

Yes. AI-based image processing can detect defects, deviations and unusual features on products or surfaces.

There are different approaches. In supervised defect detection, the AI learns from labelled training images which features indicate certain defects. In anomaly detection, on the other hand, it can learn from images of good parts what a defect-free state looks like and identify deviations from it.

Which technology is suitable depends on the inspection task, and as the expert in your field, you can train the AI yourself in the DENK Vision AI Hub.

Yes, AI-based image processing can be used for optical measurement. For example, objects, contours or relevant image areas can be detected and then evaluated in terms of their position, size or distances.

However, reliable measurements in physical units require a suitable calibration of the camera system. Depending on the application, perspective, resolution, lens distortion and other influencing factors must also be taken into account.

DENKnet handles, for example, the detection or segmentation of the relevant features. The actual measurement is calculated on the basis of the image data and the respective calibration.

Depending on the application, we can work with both monochrome images (greyscale images) and RGB colour images.

Which option is better suited depends on the inspection task. If, for example, colour differences need to be detected, colour images make sense. For detecting shapes, contours or certain surface features, on the other hand, greyscale images may already be sufficient.

What matters is that the relevant features are visible in the image and that the image quality matches the required inspection accuracy.

Yes. AI-based image processing can be trained to handle different lighting conditions, shadows and backgrounds.

Unlike many rule-based methods, not every image condition has to be described with fixed thresholds. Instead, the AI can learn to recognise relevant features despite visual changes.

To achieve this, the different conditions should be sufficiently represented in the training images. Whether additional lighting or other adjustments are necessary depends on the specific application.

With DENKnet, DENKweit relies on technologies that can be trained with images from real production environments instead of assuming idealised laboratory conditions.

That depends on the inspection task, the product variants and the complexity of the features to be detected.

With DENK Vision AI, training can start with as few as 15 training images. Especially for clearly defined classification tasks, getting started with just a few images can be possible.

More complex applications, different types of defects or strongly varying production conditions may require additional training images.

It is also important to distinguish between training and validation: even if an AI can be trained with a few images, its reliability must be verified with independent test images under real operating conditions.

The training time depends on the chosen AI technology, the number and resolution of the training images and the complexity of the task.

For certain image classification tasks, DENKweit offers particularly fast training methods. For suitable applications, training can take just a few minutes, and first results are often achieved within a few hours.

More demanding tasks such as object or instance segmentation may take more time.

In the DENK Vision AI Hub, users can start and manage training processes on their own without having to program their own AI models.

For an automated good/bad inspection, you first need images of the products to be inspected, which are uploaded to the DENK Vision AI Hub. They should show the relevant features under production conditions that are as realistic as possible.

You then train your DENKnet within a minute, because the DENK Vision AI Hub lets you annotate folder by folder.

For later use in production, the model is integrated into the inspection process and validated with suitable test images.

With "Just press play" we describe our approach of making the use of industrial vision AI as simple as possible.

Users should not have to develop their own AI architectures or bring extensive programming skills in order to automate a visual inspection task.

In the DENK Vision AI Hub, they can upload their images, annotate relevant features and start the training. The technical complexity of AI model development is handled by the platform.

This allows users to focus on their actual task: detecting and evaluating relevant features in their images.

Not anymore. We have moved away from complex test accounts that were time-consuming and costly for us.

In an initial meeting, we discuss your use case and your goal together and decide whether your problem can generally be solved with an image analysis AI. Not every use case can be implemented with a vision AI the way customers imagine.

We only start a joint project if we can actually solve your problem. When we start a joint project, we also deliver a successful AI deployment for you.

Yes. The industries in the menu are only examples to show where and how we already use image analysis AI successfully. Like the AI applications and customer stories, they help you understand the wide range of possible uses of our vision AI.

Since our DENKnets can recognise anything the human eye can recognise, there are essentially no limits. DENKweit's vision AI technologies are not restricted to specific industries.

What matters is the visual inspection task: if a relevant feature is visible to a human in the camera image, it can in principle be evaluated automatically with AI-based image processing.

Typical applications include quality control, object detection, classification, counting, text recognition or monitoring production processes. So just get in touch and we will look at your use case together.

New product variants or defect types can be added to existing vision AI projects with additional training images.

In the DENK Vision AI Hub, users can add new images, annotate them and retrain the AI. The updated model is then tested and can be deployed after successful validation.

This means companies do not have to develop a completely new image processing solution for every new product variant or inspection requirement.

This is particularly interesting for production companies with frequently changing products and for machine builders whose machines are meant to take on different tasks over the long term.

Yes. A vision AI platform can provide different models and technologies for various visual inspection tasks.

With the DENK Vision AI Hub, for example, applications for image classification, object detection, segmentation or text recognition can be trained and managed.

This allows companies to automate different tasks on a common technological basis, such as inspecting surfaces, checking labels or counting components.

Which models can run simultaneously on a particular system depends on the hardware used, the performance requirements and the respective integration architecture.

Machine and plant builders can integrate DENKweit's vision AI technology as a software component into their existing or newly developed systems.

A camera captures the relevant process or the product to be inspected. Customers then train the AI themselves in the DENK Vision AI Hub with a few images from real, ongoing operation. The results can be passed on, for example, to a machine control system or higher-level software.

DENKweit provides various integration options for this, including APIs, SDKs and local inference solutions.

The machine builder does not have to develop the AI technology themselves but can focus on their core competence: designing and automating their machines.

New inspection tasks can also be trained via the DENK Vision AI Hub after commissioning.

The required hardware depends on the application. In principle, you need a suitable camera for image acquisition and a system for processing the image data.

DENKweit's vision AI technology is not tied to a specific camera manufacturer or a fixed all-in-one hardware system.

Depending on the requirements, existing industrial cameras and PC-based systems can be used, for example. When selecting hardware, image resolution, processing speed, interfaces and available computing power should be taken into account in particular.

DENKweit focuses on the AI software and can recommend suitable hardware and integration partners if required.

Yes. In many cases, existing camera systems can continue to be used for AI-based image processing, including AOI systems that are already installed.

Since DENKweit offers a hardware-independent software solution, its use is not limited to cameras from a particular manufacturer.

The prerequisite is that the image data can be provided in suitable quality and via a compatible interface. Whether an existing camera is sufficient for the desired inspection task depends, among other things, on its resolution, frame rate and the visible features.

This often allows companies to keep using existing hardware investments and avoid additional purchase costs.

Hardware-independent vision AI allows machine builders to select cameras and computer systems that match their technical and economic requirements.

The AI technology is not necessarily tied to a specific camera model or a proprietary all-in-one system.

This offers several advantages: existing hardware can be reused under suitable conditions, machines can be configured more flexibly and different systems can be built on the same AI technology.

DENKweit deliberately focuses on developing vision AI software. This allows machine builders to use the hardware best suited to their application and to work with specialised integration partners if required.

Inference is the evaluation of a new image by an already trained AI model.

The inference time depends on various factors, in particular the AI technology used, the image resolution, the hardware used and the number of objects or features to be detected.

For industrial applications, it is crucial that the image evaluation takes place within the available process or cycle time.

DENKweit offers different options for integrating and running the AI models.

In individual cases, we can always make a network leaner, i.e. fine-tune it, when every millisecond counts. With appropriate GPU hardware, we have so far met every cycle time required by our customers, including in the double-digit millisecond range.

The suitable image size depends on the vision AI technology used and the respective application.

In principle, resolution and image section must be chosen so that the relevant features can be reliably detected. Very large images can place additional demands on memory, processing speed and evaluation strategy.

DENKweit also supports the evaluation of large-format image data, for example high-resolution drone or microscopy images, orthophotos or line scan camera images of up to 200 MB and more.

Which image sizes are supported for a specific model and a particular integration environment should be checked against the technical specifications.

Yes. DENKweit's vision AI technologies can also be used to evaluate image data from video recordings.

To do this, individual video frames are analysed by the AI. Depending on the application, objects, people, vehicles or certain states can be detected and classified, for example.

The evaluation can be relevant both for recorded video material and for live camera images.

Whether real-time evaluation is possible depends on the required frame rate, the AI technology and the available computing power. For applications in which objects are tracked across several frames or temporal sequences are evaluated, additional processing steps may be necessary.

With video recordings, make sure to observe the legal situation regarding personal data. Our DENK pixelation might be of interest to you in order to act in compliance with the GDPR.

DENKweit specialises in developing vision AI software and its own AI technologies.

We do not manufacture our own industrial camera systems or complete image processing machines. Instead, we develop the AI technology that can be integrated into various hardware and automation solutions.

This gives our customers flexibility in choosing cameras and computer systems.

For the implementation of complete image processing solutions, we work with specialised partners from the fields of camera technology, mechanical engineering and system integration where required.

AI-based quality control can pay off in particular when visual inspections are carried out regularly, manual inspection effort is high or defects lead to scrap, rework and additional costs.

Key factors for economic viability include production throughput, the cost of defects, the integration effort and ongoing operating costs.

Another factor is the reusability of DENK Vision AI: if additional inspection tasks can be automated with the same platform, certain development and integration costs are spread across several applications.

DENKweit therefore takes an approach that considers not only technical feasibility but also the economic benefit of a vision AI application.

No, not necessarily. Machine builders can use specialised vision AI technologies without having to develop their own AI architectures or build a complete AI development department.

With the DENK Vision AI Hub, users can train their own AI models on their image data. DENKweit takes care of the actual technology development, the training infrastructure and the further development of the AI platform.

This allows machine builders to focus on their core competencies while integrating AI-based functions into their machines.

Suitable expertise is still required for the technical integration and ongoing operation. DENKweit supports you with these tasks.

A central vision AI platform makes it easier to transfer AI-based inspection tasks to further applications and locations.

Different AI models can be trained and managed in the DENK Vision AI Hub. For new products or inspection requirements, additional training images can be added and adapted models created.

This allows companies to automate additional visual inspection processes without having to develop a completely new AI technology for every use case.

When transferring to other machines or sites, however, differences in camera perspective, lighting, hardware and production conditions must be taken into account. Before productive use, every model should be validated under the respective conditions.

Anyone who develops a separate AI solution for every inspection task often also has to build the associated tools, integrations and maintenance processes separately.

A central vision AI platform, by contrast, creates a common technological basis for different applications.

With the DENK Vision AI Hub, companies can manage various inspection tasks, train models and extend existing applications as needed.

This can reduce the effort for new AI projects and makes long-term use of the technology within a company easier.

This approach is particularly interesting (above all economically) for machine builders and industrial companies with many product variants or different production lines.

DENKweit attaches great importance to data sovereignty and the responsible handling of customer data. The server locations are in Germany.

Depending on the product, integration and agreed operating model, different options for processing and storing the data may exist. For sensitive use cases, on-premise installations or customer-owned cloud instances are also possible.

The rights to the images provided by customers remain with the respective customer.

The images are used for training and using the customer-specific vision AI applications. By providing its training platform, DENKweit does not claim any ownership rights to the uploaded images.

The specific rights of use and processing are set out in the respective contractual agreements.

No. According to our data processing approach, a customer's images are not used to train AI models for other customers. The image data provided serves the respective customer-specific use case.

This allows companies to keep control over their own image data and to use sensitive production information for training their vision AI applications. (One of the advantages compared to open-source solutions.)

Yes. DENKweit offers options to integrate and run trained AI models locally in existing systems.

Inference, i.e. the evaluation of new images by a trained model, can take place on a local computer system depending on the integration variant. This means that not every production image has to be transferred to an external cloud service for evaluation.

This is particularly interesting for industrial applications with high requirements for response times, availability or data sovereignty.

Which functions are fully available offline and which may require a connection to the training platform depends on the chosen solution.