Please use this identifier to cite or link to this item:
http://kmutnb-ir.kmutnb.ac.th/jspui/handle/123456789/574Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | RUT YATIGUL | en |
| dc.contributor | รุจ ยธิกุล | th |
| dc.contributor.advisor | WISANU JITVIRIYA | en |
| dc.contributor.advisor | วิษณุ จิตวิริยะ | th |
| dc.contributor.other | King Mongkut's University of Technology North Bangkok | en |
| dc.contributor.other | มหาวิทยาลัยเทคโนโลยีพระจอมเกล้าพระนครเหนือ | th |
| dc.date.accessioned | 2026-11-17T07:55:27Z | - |
| dc.date.available | 2026-11-17T07:55:27Z | - |
| dc.date.created | 2026 | - |
| dc.date.issued | 2026-11-17 | - |
| dc.identifier.uri | http://kmutnb-ir.kmutnb.ac.th/jspui/handle/123456789/574 | - |
| dc.description.abstract | This research investigates the effect of adaptive pooling within CNNs and attention mechanisms, specifically the Squeeze-and-Excitation (SE) network. The experiment was divided into two stages. First, SE networks were modified by replacing their original pooling mechanism with the proposed fuzzy pooling approach. Second, the same modification was applied to the main pooling layer in CNNs. The fuzzy pooling mechanism controls a weight function that determines the fusion ratio between two pooling methods using a Laplacian feature map analysis system. The approach combines min pooling and max pooling to enhance tomato ripeness classification accuracy. The method was evaluated using a tomato ripeness dataset containing five ripeness classes: 0%, 30%, 50%, 70%, and 100%. A fixed-weight mixed pooling and traditional deterministic pooling method was also included for comparative analysis. Results demonstrate that fuzzy pooling outperformed traditional methods with zero to minimal increase in computational consumption, highlighting the effectiveness of integrating logic-based adaptive pooling into CNNs and SE networks. The proposed model achieves a balance between high performance and computational efficiency. These findings contribute to the field of image classification and provide valuable insights for improving object recognition and classification tasks. | en |
| dc.language.iso | ENG | - |
| dc.publisher | King Mongkut's University of Technology North Bangkok | - |
| dc.rights | King Mongkut's University of Technology North Bangkok | - |
| dc.subject | Image Classification | en |
| dc.subject | Attention Mechanism | en |
| dc.subject | Fuzzy Pooling | en |
| dc.subject | Mixed Pooling | en |
| dc.subject | Laplacian Filter | en |
| dc.subject.classification | Computer Science:Artificial Intelligence | en |
| dc.title | Development of Deep Neural Networks for Object Detection Systems | en |
| dc.title | การพัฒนาเครือข่ายประสาทเทียมเชิงลึกสำหรับระบบตรวจจับวัตถุ | th |
| dc.type | Thesis | en |
| dc.type | วิทยานิพนธ์ | th |
| dc.contributor.coadvisor | ARAN BLATTLER | en |
| dc.contributor.coadvisor | อรัญ แบล็ทเลอร์ | th |
| dc.contributor.emailadvisor | wisanuj@kmutnb.ac.th | - |
| dc.contributor.emailcoadvisor | aranb@kmutnb.ac.th | - |
| dc.description.degreename | Master of Engineering (วศ.ม.) | en |
| dc.description.degreename | วิศวกรรมศาสตรมหาบัณฑิต (M.Eng.) | th |
| dc.description.degreelevel | Master of Engineering | en |
| dc.description.degreelevel | วิศวกรรมศาสตรมหาบัณฑิต | th |
| dc.description.degreediscipline | Production Engineering | en |
| dc.description.degreediscipline | วิศวกรรมการผลิต | th |
| Appears in Collections: | FACULTY OF ENGINEERING | |
Files in This Item:
| File | Size | Format | |
|---|---|---|---|
| s6601021810081.pdf | 3.17 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.