研究生: |
連儷晴 Li-Ching Lien |
---|---|
論文名稱: |
以遺傳演算法結合卷積神經網路搜尋最佳卷積核作影像分類之研究 A Study on Combining GA and CNN to Search Optimized Filter for Image Classification |
指導教授: |
楊傳凱
Chuan-Kai Yang |
口試委員: |
林伯慎
Bor-Shen Lin 賴源正 Yuan-Cheng Lai |
學位類別: |
碩士 Master |
系所名稱: |
管理學院 - 資訊管理系 Department of Information Management |
論文出版年: | 2020 |
畢業學年度: | 108 |
語文別: | 中文 |
論文頁數: | 125 |
中文關鍵詞: | 機器學習 、遺傳演算法 、卷積神經網路 、卷積核 、最佳化 、二維搜尋 、影像分類 |
外文關鍵詞: | Machine Learning, Genetic Algorithm, Convolution Neural Networks, Filter, Optimization, Two-dimensional searching, Image Classification |
相關次數: | 點閱:580 下載:0 |
分享至: |
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
使用卷積神經網路作影像分類時,卷積核對圖像特徵擷取,有著舉足輕重之
影響,本文提出利用遺傳演算法結合卷積神經網路,搜尋對於不同的影像資料,
具有最佳特徵擷取能力的卷積核,以進行影像分類。
本文實驗首先藉由遺傳演算法,產生不同形狀與大小的卷積核個體後,利用
適應函數結合卷積神經網路,用以評估各個體之卷積核對於影像資料的分類準確
率,再以適合度進行卷積核篩選,持續進行交配、突變演化,達成擇優演化最佳
卷積核之目的,並自動儲存到目前為止,所搜尋到的最佳卷積核模型,直到滿足
停止的條件時結束。
在實驗影像資料庫中,本文方法以遺傳演算法結合卷積神經網路,搜尋最佳
卷積核作影像分類,搜尋到之最佳卷積核,確實會優於傳統人工依憑經驗,所挑
選的正方形卷積核;據實驗結果顯示,卷積核的形狀與面積,皆會造成卷積神經
網路辨識之準確率差異;同時從實驗結果中發現,卷積神經網路與遺傳演算法結
合後,可以對不同影像資料庫,搜尋到具有較佳特徵擷取能力的卷積核,提升卷
積神經網路影像辨識之準確率;本文使用遺傳演算法結合卷積神經網路之方法,
對於較為複雜的影像資料或龐大數據資料,在優化卷積核的效能可更為明確,用
以達到節省人工調整參數的時間,並有效提升卷積神經網路之辨識準確率。
Filters of convolutional neural network have important impact on extracting features for image classification. This paper proposes a method which combines convolutional neural network (CNN) and genetic algorithm(GA), so as to search the optimized filter in various image databases for image classification.
This paper utilizes GA to generate individuals which contain filters of different shapes and sizes. Then GA combines fitness function with CNN together, in order to evaluate the fitness of each individual’s performance for image classification. According to the fitness, GA selects the individuals for the next generation, and it can achieve the purpose of optimizing filters. In the process, GA also picks the best model with optimized filters to store. The evolution process will stop, if the stop condition is satisfied.
In the experiment image databases, this paper uses GA to be combined with CNN to search for the optimized filter for image classification. Human usually rely on experience to adjust the filters, whereas the filters searched by this paper are certainly better than the square filters adjusted by human in most experiments. Outcomes show that the shapes and the sizes of the filters indeed cause the difference of accuracy for image classification. The results, such as combining GA and CNN for image classification, show that it can find the filters with stronger capability to extract features for different image databases, and increase the accuracy for image classification. Using GA and CNN can be more effective to optimize the filters, if the images are more complicated and enormous. Furthermore, it also can decrease the cost and time compared with human effort.
[1] Yann LeCun, L´eon Bottou, Yoshua Bengio, and Patrick Haffner. Gradientbased learning applied to document recognition. 1998.
[2] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural networks. In NIPS, 2012.
[3] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed,
Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. 2015 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), pages 1–9, 2014.
[4] B. M. Sherin and M. H. Supriya. Ga based selection and parameter optimization
for an svm based underwater target classifier. 2015 International Symposium
on Ocean Electronics (SYMPOL), pages 1–7, 2015.
[5] Basheer Qolomany, Majdi Maabreh, Ala I. Al-Fuqaha, Ajay Gupta, and Driss
Benhaddou. Parameters optimization of deep learning models using particle
swarm optimization. 2017 13th International Wireless Communications and
Mobile Computing Conference (IWCMC), pages 1285–1290, 2017.
[6] Jo˜ao Carlos Xavier, Alex Alves Freitas, Antonino Feitosa Neto, and
Teresa Bernarda Ludermir. A novel evolutionary algorithm for automated machine learning focusing on classifier ensembles. 2018 7th Brazilian Conference
on Intelligent Systems (BRACIS), pages 462–467, 2018.
[7] F´elix-Antoine Fortin, Fran¸cois-Michel De Rainville, Marc-Andr´e Gardner, Marc
Parizeau, and Christian Gagn´e. DEAP: Evolutionary algorithms made easy.
Journal of Machine Learning Research, 13:2171–2175, jul 2012.
[8] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep
network training by reducing internal covariate shift. ArXiv, abs/1502.03167,
2015.
[9] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew
Wojna. Rethinking the inception architecture for computer vision. 2016 IEEE
124
Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818–
2826, 2015.
[10] Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, and Alex Alemi. Inceptionv4, inception-resnet and the impact of residual connections on learning. In
AAAI, 2016.
[11] 吳明穎. 國立交通大學機構典藏之遺傳演算法, 2004.
[12] Jeffrey R. Sampson. Adaptation in natural and artificial systems (john h. holland). 1976.
[13] Yann LeCun, D Touresky, G Hinton, and T Sejnowski. A theoretical framework
for back-propagation. In Proceedings of the 1988 connectionist models summer
school, volume 1, pages 21–28. CMU, Pittsburgh, Pa: Morgan Kaufmann, 1988.
[14] Nilesh Patil. 2018 nilesh patil. powered by jekyll using the so simple theme.,
2018.
[15] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for
large-scale image recognition. CoRR, abs/1409.1556, 2014.
[16] Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. CoRR, abs/1412.6980, 2014.
[17] John C. Duchi, Elad Hazan, and Yoram Singer. Adaptive subgradient methods
for online learning and stochastic optimization. J. Mach. Learn. Res., 12:2121–
2159, 2010.
[18] Tijmen Tieleman and Geoffrey Hinton. Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural networks
for machine learning, 4(2):26–31, 2012.