Seminars

View all Seminars  |  Download ICal for this event

Learnings from Automatic Defect Recognition In Digital Radiographic Images With Deep Learning

Series: Department Seminar

Speaker: Soumendra Dhanee, Chief Scientist, Deep learning, VJ Imaging Technologies

Date/Time: Apr 16 16:00:00

Location: CSA Lecture Hall (Room No. 117, Ground Floor)

Abstract:
Non Destructive Testing and Evaluation (NDT and NDE) is a critical part of many industrial manufacturing and inspection processes, often bottlenecked by reliance on human operators. In building defect segmentation models for assisting human operators working with high energy x-ray images, we faced two key challenges: extremely small defects and lack of annotated datasets. We discuss our approach to solving these problems resulting in a background separation model with 99.4% pixel-classification accuracy and a defect recognition model with more than 80% accuracy on average across a number of defects. Our results have been accepted for publication in NDE 2023 under the title AIADRsuite: Automatic Defect Recognition In Digital Radiographic Images Of Castings With Deep Learning.

Speaker Bio:
Soumendra Dhanee has been working as Chief Scientist, deep learning, for VJ Imaging Technologies (https://vjt.com/), focussing on deep learning for computer vision - specifically image segmentation, contrastive and unsupervised pretraining, attention/transformers, and selective/structured state space models.

Host Faculty: Prof. Chiranjib Bhattacharyya