JavaScript is disabled for your browser. Some features of this site may not work without it.
| dc.contributor.author | Asaminew, Gizaw Egu | |
| dc.date.accessioned | 2021-10-19T11:17:36Z | |
| dc.date.available | 2021-10-19T11:17:36Z | |
| dc.date.issued | 2020-07-23 | |
| dc.identifier.uri | http://ir.bdu.edu.et/handle/123456789/12795 | |
| dc.description.abstract | Water bottle defect detection is the main issue in the bottling process. The quality of the water is influenced by the encapsulation quality, such as no cap, high cover, crooked (curved) cap, which could bring about serious problems. When the application torque is inconsistent, the result is inconsistent seals resulting in the scratch defect. This study is concerned with detecting broken cap, absence of cap, scratched cap, and foreign objects in the water. To hit the main goal first, we collected water bottle images from different sources. Then, we have categorized the collected images as the training, validation, and testing data, which contains 7800, 1800, and 184 images respectively. After data organization, image preprocessing such as resizing with bicubic interpolation, histogram equalization by contrast limited adaptive histogram equalization, and smoothing with bilateral filter were applied to the labeled data. This enhances the processing time and accuracy of the defect detection algorithm. Next, we developed a convolutional neural netwowrk model with six convolutional and four dense layers, that is trained on the preprocessed dataset. The Python programming language was used to implement the model on the top of the Tensorflow and Keras API, by testing and selecting the best learning rate and optimizer. Additionally, you only look once version 3 object detection algorithm was used to detect bottles from the input image. After all, we have tested the model on test dataset and the result is presented using confution matrix. Also, the accuracy of the developed model is compared with the accuracy of the pre-trained VGG16 and ResNet50 model by training them on our dataset. VGG16 has got 97% validation and 95% training accuracy while ResNet has achieved 10% validation and 96% training accuracy, but the developed model has achieved 96% training accuracy and 99% validation accuracy. | en_US |
| dc.language.iso | en_US | en_US |
| dc.subject | ELECTRICAL AND COMPUTER ENGINEERING | en_US |
| dc.title | AN AUTOMATIC WATER BOTTLE DEFECT DETECTION SYSTEM USING DEEP LEARNING APPROACH | en_US |
| dc.type | Thesis | en_US |
