| DC 欄位 |
值 |
語言 |
| DC.contributor | 生醫科學與工程學系 | zh_TW |
| DC.creator | 雷翔宇 | zh_TW |
| DC.creator | Hsiang-Yu Lei | en_US |
| dc.date.accessioned | 2021-8-9T07:39:07Z | |
| dc.date.available | 2021-8-9T07:39:07Z | |
| dc.date.issued | 2021 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=107827030 | |
| dc.contributor.department | 生醫科學與工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 本研究使用深度學習神經網路方式建立影像重建與自動辨識功能，在基於編碼與解碼(Encoder-Decoder)框架下搭建一平行式深度學習神經網路模型，將所記錄的光學繞射訊號進行影像之重建，達到具傳統光學顯微鏡20倍物鏡下影像解析度，並透過遮罩區域卷積神經網路進行物件偵測，利用邊界框鎖定目標偵測物進行定位與搭配遮罩描繪出物體的輪廓實現例項分割，再加以對目標偵測物進行分類分析。本光學繞射訊號經由建立一套新型計算成像系統所獲取，具有顯微系統微型化與大幅提升影像視野之功能，此無透鏡光學影像還原系統與演算方法利用純量繞射理論簡化了光學成像設備，免除龐大複雜的光學元件，只需由非同調光與影像傳感器組成，透過控制光源的空間相干性在傳感器上記錄光學繞射訊號，重建之顯微影像解析度在0.6um狀態下，視場範圍可高達30mm2。
最終我們使用人體血液檢體與鼻咽癌細胞進行實際細胞影像還原與辨識以驗證本研究所研製之系統與重建模型的效果。透過細胞的表徵分析、分類、定位以及自動計數，證實本研究所建立具自動辨識分析之計算成像系統將有機會取代傳統顯微鏡進行更快速有效的影像辨識與分析。 | zh_TW |
| dc.description.abstract | In this study, deep learning is used to reconstruct image and implement the object detection. A parallel deep learning neural network model, which is based on the Encoder-Decoder framework, is built to reconstruct the recorded diffraction patterns back to the images taken by a traditional microscope with a 20x objective lens. Then, Mask Region-based Convolutional Neural Network (Mask R-CNN) is used to not only implement the object detection for the reconstruction images through bounding boxes and achieve instance segmentation by generating mask but also classify each object. The diffraction patterns are recorded by a new computational imaging system which miniaturizes the microscopy system and greatly expands the field of view. This lens-free computational imaging system and calculation method simplify the optical imaging equipment through scalar diffraction theory. It eliminates the need for large and complex optical components and is only composed of light source and image sensor. By controlling the spatial coherence of the light source, the diffraction patterns are recorded at 0.6um resolution, and the field of view can be up to 30mm2.
In the end, we take human blood and Nasopharyngeal Carcinoma cancer cells as samples of reconstruction and object detection to verify the effect of the system and reconstruction model built in this study. Through implementing of cell characterization analysis, classification, detection and automatic counting, the computational imaging system of this study can be supposed to have the opportunity to replace traditional microscopes by fast and effective image identification and analysis. | en_US |
| DC.subject | 深度學習 | zh_TW |
| DC.subject | 物件偵測 | zh_TW |
| DC.subject | 例項分割 | zh_TW |
| DC.subject | 計算成像系統 | zh_TW |
| DC.subject | 血液 | zh_TW |
| DC.subject | 鼻咽癌細胞 | zh_TW |
| DC.subject | deep learning | en_US |
| DC.subject | object detection | en_US |
| DC.subject | instance segmentation | en_US |
| DC.subject | computational imaging system | en_US |
| DC.subject | human blood | en_US |
| DC.subject | Nasopharyngeal Carcinoma cancer cell | en_US |
| DC.title | 深度學習計算成像系統於生物醫學顯微影像之重建與分析 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Deep Learning-based Computational Imaging System for Reconstruction and Analysis of Biomedical Microscopy Image | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |