Deep learning based approach for water crystal classification

VIETNAM NATIONAL UNIVERSITY, HANOI UNIVERSITY OF ENGINEERING AND TECHNOLOGY DOAN THI HIEN DEEP LEARNING-BASED APPROACH FOR WATER CRYSTAL CLASSIFICATION MASTER THESIS Major: Computer Science HA NOI - 2021 Abstract Almost the earth’s surface area is covered by water As it is pointed out in the 2020 edition of the World Water Development Report, climate change challenges the sustain- ability of water resources It is important to monitor the quality of water to preserve sustainable water resources Quality of water can be related to the water crystal struc- ture, solid-state of water, methods to understand water crystal help to improve water quality First step, water crystal exploratory analysis has been initiated under cooper- ation with the Emoto Peace Project (EPP) The 5K EPP Dataset has been created as the first world-wide small dataset of water crystals Our research focused on reducing inherent limitations when fitting machine learning models to the 5K EPP dataset One major result is the classification of water crystals and how to split our small dataset into most related groups Using the 5K EPP dataset human observations and past researches on snow crystal classification, we provided a simple set of visual labels to name water crystal shapes, with 12 categories A deep learningbased method has been used to auto- matically the classification task with a subset of the labeled dataset The classification achieved high accuracy when fine-tuning the ResNet pretrained model Keywords: Water crystal, Deep learning, Fine-tuning, Supervised, Classification iii Acknowledgements I would first like to thank my thesis supervisor Dr Tran Quoc Long, Head of the Depart- ment of Computer Science at the University of Engineering and Technology Thanks for his insightful comments both in my work and in this thesis, for his support, and many motivating discussions I also want to acknowledge my co-supervisor Dr Frederic Andres from the National Institute of Informatics, Japan for offering me the internship opportunities at NII, Japan and leading me working on diverse exciting projects Without his support and experience, I could not achieve today result Besides, I have been very privileged to get to know and to collaborate with many other great collaborators Finally, I must express my very profound gratitude to my family for providing me with unfailing support and continuous encouragement throughout my years of study and through the process of researching and writing this thesis This accomplishment would not have been possible without them iv Declaration I declare that the thesis has been composed by myself and that the work has not be submitted for any other degree or professional qualification I confirm that the work submitted is my own, except where work which has formed part of jointly-authored publications has been included My contribution and those of the other authors to this work have been explicitly indicated below I confirm that appropriate credit has been given within this thesis where reference has been made to the work of others This study was conceived by all of the authors I carried out the main idea(s) and implemented all the model(s) and material(s) I certify that, to the best of my knowledge, my thesis does not infringe upon anyone’s copyright nor violate any proprietary rights and that any ideas, techniques, quota- tions, or any other material from the work of other people included in my thesis, pub- lished or otherwise, are fully acknowledged in accordance with the standard referencing practices Furthermore, to the extent that I have included copyrighted material, I certify that I have obtained a written permission from the copyright owner(s) to include such material(s) in my thesis and have fully authorship to improve these materials Master student Doan Thi Hien v Table of Contents Abstract .iii Acknowledgements .iv Declaration v Table of Contents .vi Acronyms viii List of Figures x List of Tables .xi Introduction 1.1 Motivation 1.2 Problem Statement 1.3 Difficulties and Challenges 1.4 Common Approaches 1.5 Contributions and Structure of the Thesis .6 Related Work 2.1 Manually Approaches 2.2 Deep Learning-Based Approaches The 5K EPP dataset .11 3.1 Data collection .11 3.2 Water crystal definition 12 Materials and Methods 17 4.1 Theoretical Basis 17 4.1.1 Convolutional Neural Network 17 vi 4.1.2 Convolutional Autoencoder .19 4.1.3 Residual Connection 20 4.2 Overview of Proposed System .20 4.3 Unsupervised Learning 21 4.3.1 Residual Autoencoder Model .21 4.3.2 K-means algorithm .23 4.4 Supervised Learning 24 4.5 Data processing 26 4.5.1 Background removing 26 4.5.2 Dataset diversity 27 4.5.3 Imbalanced data 27 Experiments and Results .29 5.1 Implementation and Configurations 29 5.1.1 Model Implementation .29 5.1.2 Training and Testing Environment 30 5.2 Datasets and Evaluation methods 31 5.2.1 Dataset 31 5.2.2 Metrics and Evaluation .32 5.3 Performance of Proposed model 33 5.3.1 Residual Autoencoder model (RAE) 33 5.3.2 K-means for Clustering 35 5.3.3 Training Classification Model 36 Conclusions 40 References 41 vii Acronyms 2D 2-Dimensional 3D 3-Dimensional Adam Adaptive Moment Estimation AI Artificial Intelligence BCE Binary Cross Entropy CAE Convolutional Auto Encoder CNN Convolutional Neural Network CPU DNN Central Processing Unit Deep Neural Network EPP Emoto Peace Project FC Fully Connected GPU Graphics Processing Unit ILSVRC ImageNet Large Scale Visual Recognition Challenge MASC Multi-Angle Snowflake Camera MLP Multilayer Perceptron viii RAE Residual Auto Encoder ix ReLU Rectified Linear Unit RNN Recurrent Neural Network SGD Stochastic Gradient Descent SSIM Structural Similarity Index x ... of water to preserve sustainable water resources Quality of water can be related to the water crystal struc- ture, solid-state of water, methods to understand water crystal help to improve water. .. a system to classify water crystal based on deep learning methods We are interested in applying a deep learning model to extract the high-meaning features from 2D water crystal images then use... the classification tasks The most popular approach to classify crystal is manually classification, which is based on human observation In recent years, with the advent of deep learning, deep learning- based