臺灣博碩士論文加值系統

本研究使用Triplet Loss處理分類問題，訓練模型學習比對兩張圖片的車型相似度，此方法有著新車型出現時不需重新訓練的優勢，可藉此降低訓練成本。此外，為了進一步提高比對的準確率，加入了下列幾項機制：1.為了減少車窗、車身金屬受環境光源影響，在比對時造成干擾，使用物件偵測模型取出和車型較相關的區域進行比對。2.為了減少比對的圖片數量，加快系統執行速度，藉由辨識車輛的廠牌，限縮要比對的車型圖片。3.對於車型外觀差異可能只有局部，透過融合各區域的比對結果，進一步提高準確率。4.開放式的圖庫設計，維護人員可透過更新、修正圖片資訊，維持系統的準確率。最後經由實驗，以362個車型類別資料訓練後，針對不在訓練集中，屬於開集合的93個車型類別進行辨識，根據實驗結果證明，其Rank3準確率可達到85%。

摘要
誌謝
目錄
表目錄
圖目錄
第一章 緒論
1.1 研究背景和動機
1.2 研究目的
1.3 論文架構
1.4 名詞解釋
第二章 文獻探討
2.1 物件偵測（Object Detection）
2.2 物件分類（Object Classification）
2.2.1 ResNet
2.2.2 DenseNet
2.3 車輛識別
2.3.1 車輛重新識別 (Vehicle Re-Identification)
2.3.2 車輛廠牌車型識別 (Vehicle Make And Model Recognition)
第三章 研究內容與方法
3.1 人工智慧輔助刊登系統
3.2 AI辨識流程
3.2.1 車型關鍵區域偵測
3.2.2 車型關鍵區域圖片比對
3.2.3 車型關鍵區域比對圖庫限縮
3.2.4 車型關鍵區域比對結果分數融合
3.2.5 車型關鍵區域圖庫資料淨化
3.3 深度學習模型訓練資料來源
第四章 實驗結果
4.1 實驗數據資料
4.2 評估方法
4.3 超參數設定
4.4 實驗環境
4.5 實驗結果
4.5.1 模型超參數設定
4.5.2 比對區域數設定
4.5.3 比對圖片數設定
4.5.4 資料淨化結果
4.6 系統比較結果
4.6.1 模型比對所需時間
4.6.2 成功案例
4.6.3 分數融合後成功案例
4.6.4 失敗案例
第五章 結論
參考文獻

[1]Redmon, J., et al. You Only Look Once: Unified, Real-Time Object Detection. 2015. arXiv:1506.02640 DOI: 10.48550/arXiv.1506.02640.
[2]Bochkovskiy, A., C.-Y. Wang, and H.-Y.M. Liao YOLOv4: Optimal Speed and Accuracy of Object Detection. 2020. arXiv:2004.10934 DOI: 10.48550/arXiv.2004.10934.
[3]Redmon, J. and A. Farhadi YOLOv3: An Incremental Improvement. 2018. arXiv:1804.02767 DOI: 10.48550/arXiv.1804.02767.
[4]Wang, C.-Y., A. Bochkovskiy, and H.-Y.M. Liao YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. 2022. arXiv:2207.02696 DOI: 10.48550/arXiv.2207.02696.
[5]He, K., et al. Deep Residual Learning for Image Recognition. 2015. arXiv:1512.03385 DOI: 10.48550/arXiv.1512.03385.
[6]Krause, J., et al., 3D Object Representations for Fine-Grained Categorization. 2013 IEEE International Conference on Computer Vision Workshops, 2013: p. 554-561.
[7]Liu, Y. Car-Recognition. 2018; Available from: https://github.com/foamliu/Car-Recognition.
[8]Huang, G., et al. Densely Connected Convolutional Networks. 2016. arXiv:1608.06993 DOI: 10.48550/arXiv.1608.06993.
[9]Liu, H.Y., et al., Deep Relative Distance Learning: Tell the Difference Between Similar Vehicles. 2016 Ieee Conference on Computer Vision and Pattern Recognition (Cvpr), 2016: p. 2167-2175.
[10]Marín-Reyes, P.A., et al., Unsupervised Vehicle Re-Identification using Triplet Networks. 2018.
[11]Zhang, C., et al., A car-face region-based image retrieval method with attention of SIFT features. Multimedia Tools and Applications, 2016. 76(8): p. 10939-10958.
[12]He, B., et al., Part-regularized Near-duplicate Vehicle Re-identification. 2019 Ieee/Cvf Conference on Computer Vision and Pattern Recognition (Cvpr 2019), 2019: p. 3992-4000.
[13]Khorramshahi, P., et al. A Dual-Path Model With Adaptive Attention For Vehicle Re-Identification. 2019. arXiv:1905.03397 DOI: 10.48550/arXiv.1905.03397.
[14]Meng, D., et al., Fine-grained Feature Alignment with Part Perspective Transformation for Vehicle ReID. Proceedings of the 28th ACM International Conference on Multimedia, 2020.
[15]Zheng, Z., et al., VehicleNet: Learning Robust Visual Representation for Vehicle Re-identification. 2020. p. arXiv:2004.06305.
[16]Ma, X.R. and A. Boukerche, An AI-based Visual Attention Model for Vehicle Make and Model Recognition. 2020 Ieee Symposium on Computers and Communications (Iscc), 2020: p. 436-441.
[17]Yu, Y., et al., CAM: A fine-grained vehicle model recognition method based on visual attention model. Image and Vision Computing, 2020. 104: p. 104027.
[18]Amirkhani, A. and A.H. Barshooi, DeepCar 5.0: Vehicle Make and Model Recognition Under Challenging Conditions. IEEE Transactions on Intelligent Transportation Systems, 2023. 24(1): p. 541-553.
[19]ABC 好車網. Available from: https://www.abccar.com.tw.
[20]Schroff, F., D. Kalenichenko, and J. Philbin FaceNet: A Unified Embedding for Face Recognition and Clustering. 2015. arXiv:1503.03832 DOI: 10.48550/arXiv.1503.03832.
[21]DeepCar5.0. Available from: https://github.com/DeepCar/DeepCar5.0.
[22]Buslaev, A., et al., Albumentations: Fast and Flexible Image Augmentations. Information, 2020. 11(2): p. 125.
[23]Lee, H.-y.; Available from: https://www.youtube.com/@HungyiLeeNTU.
[24]Zhang, K., et al., Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks. IEEE Signal Processing Letters, 2016. 23: p. 1499-1503.