The doctoral dissertation in the field of Computer Science will be examined at the Faculty of Science, Forestry and Technology, Kuopio Campus.
What is the topic of your doctoral research? Why is it important to study the topic?
My doctoral research is centered on the development and optimization of methods to detect, classify, and segment defects on product surfaces using advanced deep neural network architectures. More specifically, I focus on designing lightweight neural network models tailored for real-time industrial applications. The primary goal is to make these models efficient, accurate, and computationally inexpensive. To support this, I've developed a dataset of labeled images highlighting defects on wood surfaces, which serve as the foundation for the training and evaluation of the proposed methods.
What are the key findings or observations of your doctoral research?
A significant contribution of my research to the scientific community is the creation of a unique dataset containing labeled images of wood surface defects. This dataset not only paves the way for progress in defect detection but also encourages broader explorations in related domains. I've introduced lightweight neural network models tailored for real-life industrial use. These designs balance between computational efficiency and robustness, a balance crucial for real-time applications in various production settings. The key breakthrough lies in the simultaneous enhancement of accuracy without spiraling computational demands. For the broader public and industries, the implications are profound. Efficient defect detection translates to superior quality products across sectors like wood, textile, and car manufacturing, fostering heightened safety standards, reduced waste, and potential cost savings for everyone involved.
What are the key research methods and materials used in your doctoral research?
The process of my doctoral research centered on a recognized need in the industrial quality control sector for real-time defect detection and segmentation. This was especially relevant for automated production lines in industries such as wood, textile, medicine, and car manufacturing. A significant step was the creation of a dataset filled with labeled images of wood surface defects. These images were critical for training and refining the deep neural network models. As for the key research methods, I explored convolutional neural networks (CNNs) and specialized in techniques like depthwise separable and pointwise convolutions. To tackle defect segmentation, I used CNNs in an encoder-decoder framework. In this setup, the encoder captures the image's contextual data, and the decoder uses this to identify defects. Models underwent training, validation, and iterative adjustments to achieve desired performance and efficiency. Testing was conducted in conditions similar to industrial environments to ensure practical application. The main material supporting my research was the dataset of labeled images showcasing defects on wood surfaces. This dataset was not only a training asset but also a measure to evaluate the neural network's effectiveness.
The doctoral dissertation of Mazhar Mohsin, MSc, entitled Deep learning based real-time defect detection, classification, and segmentation methods for industrial quality control will be examined at the Faculty of Science, Forestry and Technology, Kuopio Campus. The opponent will be Associate Professor Miguel Bordallo López, University of Oulu, and the custos will be Professor Pekka Toivanen, University of Eastern Finland. Language of the public defence is English.
For more information, please contact:
Mazhar Mohsin, firstname.lastname@example.org, tel. 040 836 8023