Abstract
Digital image processing is a crucial task in various companies nowadays. In rough environments, the camera lens is getting polluted over time and therefore the following image processing cannot be done properly. To protect the camera lens, a thin layer is often placed in front of the lens to prevent it from becoming dirty. But this layer must be cleaned over time to get clear pictures for the subsequent image processing. The task is to detect when the layer is too dirty and needs to be cleaned or replaced. Therefore, the images which are taken by the camera are used.
Additionally, the detection should work in different environments. Therefore, a lot of data for training is needed. To overcome this problem, a simulator is written, that can simulate the behavior of a polluted image. The simulator is used to generate a lot of training data. Afterward, a convolutional neural network is used to learn if an image is dirty or clean. This CNN is then applied to real-world data. The results show, that in general, it is possible to detect polluted images with simulated polluted images. However, the choice of pollution parameters in the simulator should be adapted to the real pollution that occurs.
Additionally, the detection should work in different environments. Therefore, a lot of data for training is needed. To overcome this problem, a simulator is written, that can simulate the behavior of a polluted image. The simulator is used to generate a lot of training data. Afterward, a convolutional neural network is used to learn if an image is dirty or clean. This CNN is then applied to real-world data. The results show, that in general, it is possible to detect polluted images with simulated polluted images. However, the choice of pollution parameters in the simulator should be adapted to the real pollution that occurs.
Original language | English |
---|---|
Qualification | Master of Science |
Supervisors/Advisors |
|
Award date | 9 Mar 2023 |
Publication status | Published - 9 Mar 2023 |
Externally published | Yes |
Keywords
- CNN; machine learning; neural network; image classification; contaminations; artificial intelligence