Identification of circuits and graphs via Artificial Intelligence and Machine Learning
keywords CIRCUITS, MACHINE LEARNING, OCR, PATTERN RECOGNITION
Reference persons STEFANO GRIVET TALOCIA
Research Groups EMC Group (Electromagnetic Compatibility)
Description The automated recognition of shapes and objects from images or videos is a key enabling factor in several application fields. Examples are basic text recognition from scanned images, or even autonomous vehicles driving. The latter is possible thanks to the identification of obstacles, people, surrounding vehicles and traffic lights from image streams coming from cameras and radars.
The objective of this thesis is the application of Artificial Intelligence and Machine Learning techniques for the automated recognition of electric circuits starting from images (a picture, a scan of a textbook, or even a sketch generated by hand). In particular, the main challenge is the identification of the circuit topology in terms of nodes, elements and connections through recognition of circuit element symbols and junctions of the corresponding connection lines. Further, the recognition of numerical data (element values and corresponding units) will allow the construction of an abstract representation of the circuit in view of its automated solution through formal methods.
Methodologies to be used and skills to be developed include:
- image processing and machine learning/pattern recognition, in order to recognize which circuit elements are included, and what is the circuit topology. In this phase, the student is expected to learn and use existing and well-established libraries and frameworks for object recognition (YOLO, RCNN), as well and tracking algorithms for identification of connections between elements hence circuit topology.
- text recognition, in order to identify component names/values and problem statement (existing libraries for OCR will be investigated and used). This second task will be developed only upon successful completion of the first object recognition task.
- circuit equation formulation using the Modified Nodal Analysis (MNA) framework and solution. Code for this task is already available from the Supervisor (no development required from the student)
- automated generation of a "clean" circuit schematic and problem solution. Code for this task is already available from the Supervisor (no development required from the student)
Required skills Good ability in mathematical formulation of problems and algorithms. Programming skills in Matlab and especially Python are required for proficient use of existing software libraries. an appropriate hardware/software infrastructure will be available for development, training, validation and inference.
Deadline 07/09/2023 PROPONI LA TUA CANDIDATURA