University of Twente Student Theses


Hybrid Learning for Leakage Detection in Sealed Detergent Containers using IR-Thermography

Ravi Prame, Akash (2021) Hybrid Learning for Leakage Detection in Sealed Detergent Containers using IR-Thermography.

[img] PDF
Abstract:Quality inspection plays a critical role in the manufacturing industry. With the recent popularity of detergent pods, ensuring utmost product quality has become a necessity for detergent manufacturers. Due to the detergent pod containers being opaque, manual quality inspection becomes slow and infeasible. Therefore, there is a need for a non-destructive testing (NDT) technique to automatically detect fluid leakage inside sealed containers. The focus of this study is to develop a method to automatically detect the presence of leakage in sealed containers. Infrared thermography (IRT) has been applied successfully by other researchers for quality inspection in cases where the test specimen is out of direct line-of-sight. Therefore, IRT has been identified as a suitable method to capture the information required for this task. Therefore, using thermal image data, we aim to build an image classification system to distinguish between instances of leakage and non-leakage. We propose three alternate approaches for this task, namely handcrafted featurebased approach, convolutional neural network (CNN) based approach and a hybrid fusion approach combining multiple feature sources or classifiers, or both. The CNN model outperforms the baseline feature-based approach with a 4-fold accuracy of 94.48%. The two hybrid fusion schemes namely, late-fusion and early-fusion provide an improvement to the pure CNN approach with a highest overall accuracy of 95.63% obtained over a 4-fold cross validation split.
Item Type:Essay (Master)
Tembo Group B.V., Kampen, Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page