Hardware and Software Solutions for Image Analysis
From image analysis using classical algorithms to AI-based functions using neural networks, our teams design the right hardware and software for your project.
AI-based image analysis
Our AI image analysis services help you integrate AI algorithms into your vision application. Packages range from training convolutional neural networks (CNNs) to optimizing AI models and integrating them with your target application.
Image data and AI models
Our range of services include:
Data preparation:
Cleaning and preparation of data sets, including data enrichment and preprocessing steps such as normalization, feature engineering, etc.
AI model optimization:
Optimization of hyperparameters, selection of appropriate optimizers, loss functions and regularization techniques.
Use techniques such as transfer learning and fine-tuning to adapt existing AI models to specific use cases.
AI model validation:
Metrics for evaluating AI model performance (e.g., accuracy, F1 score, ROC-AUC, etc.).
Perform cross-validation: test the AI model on different data sets to ensure generalization.
Deploy and scale to target applications
Our range of services include:
Deployment of AI models:
Deploying AI models to the operational system
Integrate AI models into existing software architectures (APIs, microservices) and ensure models can respond to queries in real time or efficiently.
Monitor and maintain AI models:
Monitor AI models for consistent and accurate predictions
Adjust or retrain AI models in the event of concept drift or new data requirements.
Hardware solutions for AI image analysis
FPGAs allow image data to be processed directly on a frame grabber or embedded vision device – from image capture to analysis results. With the integration of AI accelerators and NPUs, industrial PCs can also implement computationally intensive applications using CNNs.
AI with a Basler frame grabber and VisualApplets
Graphical FPGA programming with VisualApplets can be used to integrate suitable network architectures of varying size and complexity. Network and parameter information can also be imported from third-party software and training tools such as TensorFlow.
Our range of services include:
Implementation: If you have already created and trained a mesh, we transfer and implement it together with the weights (parameters) on the frame grabber's FPGA.
Full development: We take care of the setup and training of the network as well as the entire FPGA implementation.
AI applications in embedded environments
The most compact and cost-effective AI vision systems are developed using embedded technology: Smart edge devices offer fast runtimes, low latency, and enhanced privacy and security.
Our range of services include:
Development of neural networks with different architectures, e.g. U-Net architecture, MobileNet backbone.
Adaptation and training of existing neural networks
Integration and command of AI accelerators
Integration with target systems
