Matrox Design Assistant Tools

Classification steps

Deep-learning-based image classification
Using deep learning!
Matrox Design Assistant X includes classification steps for automatically categorizing image content using machine learning. These steps make use of deep learning—specifically convolutional neural network (CNN)—technology in two distinct approaches.
 
The first approach—implemented by the CNNClassIndex step—assigns images or image regions to pre-established classes. Results for each image or image region consist of the most likely class and a score for each class. The second approach—implemented by the CNNClassMap step—generates maps indicating the pre-established class and score for all image neighborhoods. These classification steps are particularly well suited for analyzing images of highly textured, naturally varying, and acceptably deformed goods.
Users can train a CNN on their own—using the MIL CoPilot interactive environment available separately as part of Matrox Imaging Library (MIL) X software—or commission Matrox Imaging to do so using previously-collected images that are both adequate in number and representative of the expected application conditions. Different types of training, such as transfer learning and fine-tuning, are supported, all starting from one of the supplied pre-defined CNN architectures. MIL CoPilot provides what is needed to build the required training dataset, including the labeling of images and augmenting the dataset with synthesized images, as well as to monitor and analyze the training process. Training is accomplished using a NIVIDIA GPU or x64-based CPU while inference is performed on a CPU in a Matrox Imaging vision controller, smart camera, or third-party computer, avoiding the need for specialized GPU hardware.

CNNClassIndex step

CNNClassIndex step

 

CNNClassMap step

CNNClassMap step