Deep Learning

With this new Aphelion extensions, there is no need to be a programmer to develop an algorithm based on Artificial Intelligence (AI). The Deep Learning extensions compatible with the Aphelion Imaging Software Suite help the user to develop an Artificial Intelligence application without having to be an expert in the field.

The first step of the development of an AI application involves collecting data to be used to train the Deep Learning system. Thanks the Annotate software product, fully compatible with the Aphelion Imaging Software Suite, highly configurable to be adapted to the specific annotation process, the user can seamlessly and quickly annotate images and videos. Annotated data are then used to train the Convolutional Neural Network (CNN).

The second step is the training process. This step is not straightforward as it requires an expertise in Deep Learning. It involves first splitting the set of annotated data into two sets: A training dataset and an evaluation dataset, second selecting the best CNN architecture, third generating the model, and fourth providing the classification result and the evaluation report. To help the user and save time, ADCIS can take in charge this task. Contact ADCIS to learn more about this offer.

The final step includes the use of the CNN model that has been generated in the previous step. The Deep Learning Processing extension provides all the tools to perform that task. The function that applied the CNN to an image is available in the Aphelion Developer Graphical User Interface or in the Aphelion SDK libraries. The function applies a model to an image, and then automatically classifies the image, detects and/or classifies all objects present in the image. It is also possible to segment an image if the model that has been generated is based on image segmentation.