AI辨識地標或景點是目前人工智慧發展中重要的一環，其重要性在於能夠提高旅遊體驗和地理資訊辨識的效率，為使用者提供更好的服務體驗。Google Cloud Vision 的地標識別API已提供許多國際知名地標圖片辨識的服務，但以台灣來說能出現在其中的也只有101大樓等屈指可數的景點，因此建立台灣在地化的景點辨識系統進而推廣台灣景點資訊有其重要性。

本研究的目的在於透過深度學習，建立一套台南景點圖片識別系統，我們希望能在即使不參考照片GPS定位只使用照片的元素(像素、物件的組合)也能辨識各種景點。本研究的景點資料集來自Google Image，移除非景點物件(人物、汽車等)占比較重的圖片，並藉由影像分割方法把少部分的非景點物件進行物件移除，盡可能保留較多的圖片且減少非景點物件的特徵對圖片識別造成的影響；接著使用多種資料擴增方法使資料量增長為原本的5倍並使每個景點的資料量保持平衡；在模型訓練的部分，我們使用了VGG16、MobileNetV2、NASNet、DenseNet共4種模型並整合集成學習的方式於圖片分類預測，經過實驗證明，本研究所提出景點辨識方法的準確率可達89%。

我們希望可以應用本研究所提出景點辨識方法，在後續的研究上把全台灣的景點都納入其中，讓世界看見台灣。

The recognition of landmarks or tourist attractions by AI is an important aspect of the current development of artificial intelligence. Its importance lies in the ability to enhance the travel experience and improve the efficiency of geographic information recognition, providing users with a better service experience. The Google Cloud Vision API has provided many internationally renowned landmark image recognition services, however, in Taiwan, only a handful of attractions such as the Taipei 101 can appear in them, therefore it is important to establish a localized attraction identification system in Taiwan to promote Taiwan's scenic spot information.

This study aims to recognize various tourist attractions by deep learning, solely based on the elements of photos (pixels and combinations of objects) without relying on photo GPS positioning. The dataset of tourist attractions was obtained through Google Images, which removed the pictures of heavy (more than 25%) non-attraction objects (people, cars, etc.), and for the remaining pictures, we performed data clean operations by image segmentation. Furthermore, a variety of data augmentation methods were used to increase the data by 5 times and maintain a balance of each data catalogue. Finally, in the model training, we combined VGG16, MobileNetV2, NASNet, DenseNet with ensemble learning technique for the image classification. The experimental results show that the accuracy of the proposed attraction recognition method can reach 89%, which demonstrates the efficiency of this study.

In the future, the proposed tourist attraction recognition method can be expanded to include all of Taiwan's landmarks to showcase the beauty of Taiwan to the world.

摘要....i
Abstract....ii
誌謝....iii
目錄....iv
表目錄....vi
圖目錄....vii
第一章 緒論....1
1.1 研究動機....1
1.2 研究回顧....2
1.2.1 台南景點的重要性....2
1.2.2 人工智慧....3
1.3 研究目的....3
1.4論文架構....4
第二章 文獻探討....5
2.1 圖片分類介紹....5
2.1.1 地標分類的重要性....5
2.2 深度學習....6
2.3 VGG16....7
2.4 MobileNetV2....8
2.5 NASNet....9
2.6 DenseNet....10
2.7 集成學習....12
2.7.1 Bagging....12
2.7.2 Boosting....13
2.7.3 Stacking....14
2.7.4 Averaging....15
2.7.5 Voting....15
第三章 研究方法....15
3.1 研究流程圖....15
3.2 原始資料集....16
3.3 影像分割....18
3.4 人物移除....22
3.5 資料擴增....24
3.6 模型訓練....26
3.7 Ensemble Learning – Voting....27
第四章 研究結果....28
4.1 實驗設備....28
4.2 效能評估方式....28
4.2.1 準確率....29
4.2.2 精確率....29
4.2.3 召回率....29
4.2.4 F0.5 – Score....30
4.2.5 Validation GAP....30
4.3 實驗結果....30
4.4 辨識準確度較低的景點案例探討....33
4.5 模型訓練成本比較....35
第五章 結論....36
5.1 結論....36
5.2 未來展望....36
參考文獻....37
附錄....39

[1] " Google Landmark Recognition 2020~2021 Competition “, https://www.kaggle.com/competitions/landmark-recognition-2021
[2]"xctravel",https://xctravel.tw/%E5%8F%B0%E7%81%A3%E8%A7%80%E5%85%89%E6%97%85%E9%81%8A/
[3] "台灣觀光局",
https://admin.taiwan.Net.tw/FileUploadCategoryListC003330.aspx?CategoryID=b69f0538-b028-4376-bdaa-7bac74948324&appname=FileUploadCategoryListC003330
[4] X.-B. Nguyen, D. T. Bui, C. N. Duong, T. D. Bui, and K. Luu, "Clusformer: A transformer based clustering approach to unsupervised large-scale face and visual landmark recognition," in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10847-10856, 2021.
[5] C. Henkel and P. Singer, "Supporting large-scale image recognition with out-of-domain samples," arXiv preprint arXiv:2010.01650, 2020.

[6] Y. Xu, J. Li, S. Chen, S. Wang and Q. Zheng, "Autonomous driving with deep learning: A survey of state-of-art technologies," in IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 7, pp. 2870-2893, 2020.
[7] K. Narin, C. Kaya, O. Pamukcu and S. S. Cirik, "Current limitations to identify COVID-19 using artificial intelligence with chest X-ray imaging," in IEEE Access, vol. 8 , pp. 114160-114171, 2020.
[8] L. Zhang, J. Wang, X. Lu, and Y. Xue, "Two-stage discriminative re-ranking for large-scale landmark retrieval," IEEE Transactions on Multimedia, vol. 22, no. 11, pp. 2927-2940, 2020.
[9] P. Pandey, A. Kumar, A. Mittal, and S. K. Gupta, "An accurate retrieval through R-MAC+ descriptors for landmark recognition," Multimedia Tools and Applications, vol. 79, no. 35, pp. 25939-25959, 2020.
[10] Ed Sperling,"Deep Learning Spreads" , https://semiengineering.com/deep-learning-spreads/
[11] M. Zeiler and R. Fergus, "Visualizing and understanding convolutional networks," in Proceedings of the European Conference on Computer Vision, Zurich, Switzerland, Sep. 6-12, 2014, pp. 818-833.

[12] N. Zhang, X. Liu, H. Liu, Y. Fu and Z. Ji, "A novel CapsNet neural network based on MobileNetV2 structure for robot image classification," Frontiers in Neurorobotics, vol. 15, pp. 756093, 2021.

[13]J. Zhang, C. Wu, X. Yu, and X. Lei, "A novel DenseNet Generative Adversarial Network for Heterogenous low-Light image enhancement," Frontiers in Neurorobotics, vol. 15, p. 700011, 2021.
[14] B. Zoph and Q. V. Le, "Learning transferable architectures for scalable image recognition," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, Jun. 18-22, 2018, pp. 8697-8710.
[15]F. Saxen, P. Werner, S. Handrich, E. Othman, L. Dinges, and A. Al-Hamadi, "Face attribute detection with mobileNetv2 and nasNet-mobile," in 2019 11th International Symposium on Image and Signal Processing and Analysis (ISPA), 2019: IEEE, pp. 176-180.
[16]G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, "Densely connected convolutional Networks," in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4700-4708, 2017.
[17] H. Zhang, W. Mao, Y. Zhang, X. Zhang, and Z. Liu, "DenseNet-Based Deep Learning for Brain Tumor Segmentation," in Frontiers in Neuroscience, vol. 13, Mar. 2019, pp. 1-9.
[18] M. A. Ganaie, M. Hu, A. Malik, M. Tanveer, and P. Suganthan, "Ensemble deep learning: A review," Engineering Applications of Artificial Intelligence, vol. 115, p. 105151, 2022.
[19] Y. Zhang, J. Liu, and W. Shen, "A review of ensemble learning algorithms used in remote sensing applications," Applied Sciences, vol. 12, no. 17, p. 8654, 2022.
[20] A. Abbasi, A. R. Javed, F. Iqbal, Z. Jalil, T. R. Gadekallu, and N. Kryvinska, "Authorship identification using ensemble learning," Scientific reports, vol. 12, no. 1, p. 9537, 2022.
[21] L. Breiman, "Random forests," Machine learning, vol. 45, pp. 5-32, 2001.
[22] T. Chen and C. Guestrin, "Xgboost: A scalable tree boosting system," in Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, 2016, pp. 785-794.
[22] "台南旅遊網",https://www.twtainan.Net/zh-tw/attractions
[23] H. Zhang, M. Xiao, Z. Lin, J. Luo, and Q. Wu, "DenseNet-Based Deep Learning for Brain Tumor Segmentation," in Frontiers in Neurology, vol. 11, Aug. 2020, pp. 757.
[24]T.-Y. Lin et al., "Microsoft coco: Common objects in context," in Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, 2014: Springer, pp. 740-755.
[25]F. Chollet, "Xception: Deep learning with depthwise separable convolutions," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1251-1258.
[26] Hung-yiLee , "【機器學習2021】神經網路壓縮 (Network Compression) (二) - 從各種不同的面向來壓縮神經網路",https://www.youtube.com/watch?v=xrlbLPaq_Og&t=2620s&ab_channel=Hung-yiLee
[27]K. Nazeri, E. Ng, T. Joseph, F. Z. Qureshi, and M. Ebrahimi, "Edgeconnect: Generative image inpainting with adversarial edge learning," arXiv preprint arXiv:1901.00212, 2019.
[28]J. Yoo, Y. Uh, S. Chun, B. Kang, and J.-W. Ha, "Photorealistic style transfer via wavelet transforms," in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 9036-9045.
[29]F. Krüger, "Activity, context, and plan recognition with computational causal behaviour models," University, 2016.