異常事件可能造成經濟損失與人員傷亡，能夠在提早偵測出這些異常事件是
避免重大損失的唯一方法。在我們的論文中，我們提出了一種使用攝像機的及時
偵測異常事件的方法，該方法可以工作在不同環境。異常事件通常只出現在影像
的一小塊區域，因此我們使用兩階段架構來提高準確度。在第一階段，我們使用 空間時序網路從影片中找出異常的區域，在第二階段我們使用 Resnet 判斷該區域
是否為正常狀態。我們的實驗結果證明，我們提出的方法比其他方法具有更好的
準確度。

Abnormaleventscancauseeconomiclossesandcasualties. Beingabletodetectabnormal events in the early stage is the only way to avoid significant losses. In our thesis, we present a method of using cameras to detect abnormal events in time, which can work in different environments. Abnormal events usually only appear in a small area of the image,soweuseatwo­stagearchitecturetoimproveaccuracy. Inthefirstphase,weuse the spatio­temporal network to find the area that might be a abnormal events from the video. Inthesecondstage,WeuseResnettodeterminewhethertheareaisnormalornot. Our experimental results show that the proposed method has better accuracy than other methods.

[1] H.J.G.Haynes.,“Fire loss in the united states during2015,”2016.
[2] Thou­Ho Chen, Ping­Hsueh Wu, and Yung­Chuen Chiou, “An early fire­ detection method based on image processing,”vol.3,pp.1707–1710Vol.3,Oct2004.
[3] G.Marbach,M.Loepfe,andT.Brupbacher,“An image processing technique for fire detection in video images,” Fire Safety Journal, vol. 41, no. 4, pp. 285 – 289, 2006. 13th International Conference on Automatic Fire Detection,Duisburg,Germany.
[4] B.U.Töreyin,Y.Dedeoğlu,U.Güdükbay,andA.E.Çetin,“Computer vision basedmethod for real time fire and flame detection,”PatternRecognitionLetters,vol.27,no.1,pp.49–58,2006.
[5] K.Muhammad,J.Ahmad,Z.Lv,P.Bellavista,P.Yang,andS.W.Baik,“Efficient deep cnn­basedfire detection and localization in video surveillance applications,” IEEE Transactions on Systems, Man, and Cybernetics: Systems,vol.49,pp.1419–1434,July2019.
[6] A.J.DunningsandT.P.Breckon,“Experiment ally defined convolutional neural network architecture variants for non­temporal real­time fire detection,”pp.1558–1562,Oct2018.
[7] G.Huang,Z.Liu,L.v.d.Maaten,andK.Q.Weinberger,“Densely connected convolutional networks,” pp.2261–2269,July2017.
[8] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” pp. 770–778, June2016.
[9] R. K. Srivastava, K. Greff, and J. Schmidhuber, “Highway networks,” 2015. cite arxiv: 1505.00387Comment: 6 pages, 2 figures. Presented at ICML 2015 Deep Learning workshop. Full paperisatarXiv:1507.06228.
[10] D. Chen, W. Zhang, X. Xu, and X. Xing, “Deep networks with stochastic depth for acoustic modelling,”pp.1–4,Dec2016.
[11] Y. Wen, K. Zhang, Z. Li, and Y. Qiao, “A discriminative feature learning approach for deep face recognition,”2016.
[12] D. Y. T. Chino, L. P. S. Avalhais, J. F. R. Jr., and A. J. M. Traina, “Bowfire: Detection of fire in still images by integrating pixel color and texture analysis,”CoRR,vol.abs/1506.03495,2015.