Signature Algorithm

The core component of the solutions provided by IOIMACHINES is a novel image analysis algorithm invented by the founder of the company. The algorithm runs several feature detection techniques on a given set of images and selects the most successful one for the given problem. Two of these feature detection techniques are invented by the company and they show superior performance to conventional techniques especially in detecting small deviations in surface appearance from normal cases.

A machine learning technique is then used on the generated features where a neural network is trained to classify the feature vectors into predefined classes. Case studies show that the proprietary algorithm produces 100% accuracy in detecting small defects on surfaces even when they are obscured by shadows and other normal artifacts. Other case studies showed that the algorithm is powerful in detecting different types of defects on surfaces with highly varying background structures such as the inner surface of the wind turbine blade.

The IP core is delivered in an easy to use Graphical User Interface together with functions to

• A Simple GUI to let the user by muse clicks to:
  • Create new classification categories
  • Define Regions of interest in selected images
  • Assign images to defined classes for training, test and validation
  • Add new images to the training, test and validation database
  • Enter data of defects on each image
  • Create directories including different defect types
  • Generate feature vectors using a selected technique
  • Train the Neural Network
  • Test and validate the Neural network