| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 林岱鋒 | zh_TW |
| DC.creator | Tai-Feng Lin | en_US |
| dc.date.accessioned | 2021-8-4T07:39:07Z | |
| dc.date.available | 2021-8-4T07:39:07Z | |
| dc.date.issued | 2021 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=108522023 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 在臺灣傳統魚市場中有成千上萬的魚類品種，普通消費者很難準確辨別魚的種類，但目前的魚類辨識系統大多是針對非食用魚類較多，且魚類種類數量非常稀少，因此本研究針對臺灣魚市場常見之魚類，設計一個多類別數的魚類辨識系統，以提供大眾消費者藉以在魚市場識別、採購魚類產品。首先透過物件偵測找出影像中的魚體，並設計一個混合神經網路分類器，起初透過ResNet50進行分類並輸出Top-5的相似類別，接著進行第二階段模板匹配，利用孿生神經網路模型將已切割的影像與所有相似類別的模板圖像進行比對，輸出影像間的相似值，最後比較各類的相似值，輸出最大相似值類別作為最終決策。實驗結果顯示混合神經網路分類器能達到97.5%的辨識率，優於ResNet50的90.5%和孿生神經網路的89%。 | zh_TW |
| dc.description.abstract | In Taiwan’s traditional fish market, there are thousands of fish species. It is difficult for ordinary consumers to accurately identify the types of fish. However, most of the current fish identification systems are aimed at more non-edible fish, and the number of fish species is very scarce. The study designed a multi-category fish identification system for common fish in Taiwan’s fish market to provide consumers with a way to identify and purchase fish products in the fish market. First, find the fish in the image through object detection, and design a hybrid neural network classifier. At first, it classifies through ResNet50 and outputs similar categories of Top-5, and then performs the second stage of template matching, using the Siamese neural network model compares the cut image with the template images of all similar categories,
outputs the similarity values between the images, and finally compares the similarity values of various types, and outputs the largest similarity value category as the final decision. The experimental results show that the hybrid neural network classifier can achieve a recognition
rate of 97.5%, which is better than the 90.5% of ResNet50 and 89% of the Siamese neural network. | en_US |
| DC.subject | 深度學習 | zh_TW |
| DC.subject | 物件偵測 | zh_TW |
| DC.subject | 影像辨識 | zh_TW |
| DC.title | 混合深度神經網路的市場魚種辨識 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Hybrid Deep Neural Network Classifier for Market Fish Species Recognition | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |