: Segmentation of lung regions is of key importance for the automatic analysis of Chest X-Ray (CXR) images, which have a vital role in the detection of various pulmonary diseases. Precise identification of lung regions is the basic prerequisite for disease diagnosis and treatment planning. However, achieving precise lung segmentation poses significant challenges due to factors such as variations in anatomical shape and size, the presence of strong edges at the rib cage and clavicle, and overlapping anatomical structures resulting from diverse diseases. Although commonly considered as the de-facto standard in medical image segmentation, the convolutional UNet architecture and its variants fall short in addressing these challenges, primarily due to the limited ability to model long-range dependencies between image features. While vision transformers equipped with self-attention mechanisms excel at capturing long-range relationships, either a coarse-grained global self-attention or a fine-grained local self-attention is typically adopted for segmentation tasks on high-resolution images to alleviate quadratic computational cost at the expense of performance loss. This paper introduces a focal modulation UNet model (FMN-UNet) to enhance segmentation performance by effectively aggregating fine-grained local and coarse-grained global relations at a reasonable computational cost. FMN-UNet first encodes CXR images via a convolutional encoder to suppress background regions and extract latent feature maps at a relatively modest resolution. FMN-UNet then leverages global and local attention mechanisms to model contextual relationships across the images. These contextual feature maps are convolutionally decoded to produce segmentation masks. The segmentation performance of FMN-UNet is compared against state-of-the-art methods on three public CXR datasets (JSRT, Montgomery, and Shenzhen). Experiments in each dataset demonstrate the superior performance of FMN-UNet against baselines.
{"title":"Focal modulation network for lung segmentation in chest X-ray images","authors":"ŞABAN ÖZTÜRK, TOLGA ÇUKUR","doi":"10.55730/1300-0632.4031","DOIUrl":"https://doi.org/10.55730/1300-0632.4031","url":null,"abstract":": Segmentation of lung regions is of key importance for the automatic analysis of Chest X-Ray (CXR) images, which have a vital role in the detection of various pulmonary diseases. Precise identification of lung regions is the basic prerequisite for disease diagnosis and treatment planning. However, achieving precise lung segmentation poses significant challenges due to factors such as variations in anatomical shape and size, the presence of strong edges at the rib cage and clavicle, and overlapping anatomical structures resulting from diverse diseases. Although commonly considered as the de-facto standard in medical image segmentation, the convolutional UNet architecture and its variants fall short in addressing these challenges, primarily due to the limited ability to model long-range dependencies between image features. While vision transformers equipped with self-attention mechanisms excel at capturing long-range relationships, either a coarse-grained global self-attention or a fine-grained local self-attention is typically adopted for segmentation tasks on high-resolution images to alleviate quadratic computational cost at the expense of performance loss. This paper introduces a focal modulation UNet model (FMN-UNet) to enhance segmentation performance by effectively aggregating fine-grained local and coarse-grained global relations at a reasonable computational cost. FMN-UNet first encodes CXR images via a convolutional encoder to suppress background regions and extract latent feature maps at a relatively modest resolution. FMN-UNet then leverages global and local attention mechanisms to model contextual relationships across the images. These contextual feature maps are convolutionally decoded to produce segmentation masks. The segmentation performance of FMN-UNet is compared against state-of-the-art methods on three public CXR datasets (JSRT, Montgomery, and Shenzhen). Experiments in each dataset demonstrate the superior performance of FMN-UNet against baselines.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"206 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135302571","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
: The conventional approach to creating 3D surfaces from 2D medical images is the marching cube algorithm, but it often results in rough surfaces. On the other hand, B-spline curves and nonuniform rational B-splines (NURBSs) offer a smoother alternative for 3D surface reconstruction. However, NURBSs use control points (CTPs) to define the object shape and corners play an important role in defining the boundary shape as well. Thus, in order to fill the research gap in applying corner detection (CD) methods to generate the most favorable CTPs, in this paper corner points are identified to predict organ shape. However, CTPs must be in ordered coordinate pairs. This ordering problem is resolved using curve reconstruction (CR) or chain code (CC) algorithms. Existing CR methods lead to issues like holes, while some chain codes have junction-induced errors that need preprocessing. To address the above issues, a new graph neural network (GNN)-based approach named curvature and chain code-based corner detection (CCCD) is introduced that not only orders the CTPs but also removes junction errors. The goal is to improve accuracy and reliability in generating smooth surfaces. The paper fuses well-known CD methods with a curve generation technique and compares these alternative fused methods with CCCD. CCCD is also compared against other curve reconstruction techniques to establish its superiority. For validation, CCCD’s accuracy in predicting boundaries is compared with deep learning models like Polar U-Net, KiU-Net 3D, and HdenseUnet, achieving an impressive Dice score of 98.49%, even with only 39.13% boundary
{"title":"CCCD: Corner detection and curve reconstruction for improved 3D surface reconstruction from 2D medical images","authors":"MRIGANKA SARMAH, ARAMBAM NEELIMA","doi":"10.55730/1300-0632.4027","DOIUrl":"https://doi.org/10.55730/1300-0632.4027","url":null,"abstract":": The conventional approach to creating 3D surfaces from 2D medical images is the marching cube algorithm, but it often results in rough surfaces. On the other hand, B-spline curves and nonuniform rational B-splines (NURBSs) offer a smoother alternative for 3D surface reconstruction. However, NURBSs use control points (CTPs) to define the object shape and corners play an important role in defining the boundary shape as well. Thus, in order to fill the research gap in applying corner detection (CD) methods to generate the most favorable CTPs, in this paper corner points are identified to predict organ shape. However, CTPs must be in ordered coordinate pairs. This ordering problem is resolved using curve reconstruction (CR) or chain code (CC) algorithms. Existing CR methods lead to issues like holes, while some chain codes have junction-induced errors that need preprocessing. To address the above issues, a new graph neural network (GNN)-based approach named curvature and chain code-based corner detection (CCCD) is introduced that not only orders the CTPs but also removes junction errors. The goal is to improve accuracy and reliability in generating smooth surfaces. The paper fuses well-known CD methods with a curve generation technique and compares these alternative fused methods with CCCD. CCCD is also compared against other curve reconstruction techniques to establish its superiority. For validation, CCCD’s accuracy in predicting boundaries is compared with deep learning models like Polar U-Net, KiU-Net 3D, and HdenseUnet, achieving an impressive Dice score of 98.49%, even with only 39.13% boundary","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135302576","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
AHMET HİDAYET KİRAZ, FATIME OUMAR DJIBRILLAH, MEHMET EMİN YÜKSEL
{"title":"Deep feature extraction, dimensionality reduction, and classification of medical images using combined deep learning architectures, autoencoder, and multiple machine learning models","authors":"AHMET HİDAYET KİRAZ, FATIME OUMAR DJIBRILLAH, MEHMET EMİN YÜKSEL","doi":"10.55730/1300-0632.4037","DOIUrl":"https://doi.org/10.55730/1300-0632.4037","url":null,"abstract":"","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135302577","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Millions of people throughout the world suffer from the complicated and crippling condition of chronic pain. It can be brought on by several underlying disorders or injuries and is defined by chronic pain that lasts for a period exceeding three months. To better understand the brain processes behind pain and create prediction models for pain-related outcomes, machine learning is a potent technology that may be applied in Functional magnetic resonance imaging (fMRI) chronic pain research. Data (fMRI and T1-weighted images) from 76 participants has been included (30 chronic pain and 46 healthy controls). The raw data were preprocessed using fMRIprep and then parcellated using five various atlases such as MSDL, Yeo?17, Harvard, Schaefer, and Pauli. Then the functional connectivity between the parcellated Region of Interests (ROIs) has been taken as features for the machine learning classifier models using the Blood Oxygenation Level Dependent (BOLD) signals. To distinguish between those with chronic pain and healthy controls, this study used Support Vector Machines (SVM), Boosting, Bagging, convolutional neural network (CNN), XGboost, and Stochastic Gradient Descent (SDG) classifiers. The classification models use stratified shuffle split sampling to fragment the training and testing dataset during various iterations. Hyperparameter tuning was used to get the best classifier model across several combinations of parameters. The best parameters for the classifier were measured by the accuracy, sensitivity, and specificity of the model. Finally, to identify the top ROIs involved in chronic pain was unveiled by the probability-based feature importance method. The result shows that Pauli (subcortical atlas) and MSDL (cortical atlas) worked well for this chronic pain fMRI data. Boosting algorithm classified chronic pain and healthy controls with 94.35% accuracy on the data parcellated with the Pauli atlas. The top four regions contributing to this classifier model were the extended Amygdala, the Subthalamic nucleus, the Hypothalamus, and the Caudate Nucleus. Also, the fMRI data parcellated using a cortical MSDL atlas was classified using the XGboost model with an accuracy of 87.5%. Left Frontal Pole, Medial Default mode Network, right pars opercularis, dorsal anterior cingulate cortex (dACC), and Front Default mode network are the top five regions that contributed to classify the participants. These findings demonstrate that patterns of brain activity in areas associated with pain processing can be used to categorize individuals as chronic pain patients or healthy controls reliably. These discoveries may help with the identification and management of chronic pain and may pave the way for the creation of more potent tailored medicines for those who suffer from it.
{"title":"Classification of chronic pain using fMRI data: Unveiling brain activity patterns for diagnosis","authors":"REJULA V, ANITHA J, BELFIN ROBINSON","doi":"10.55730/1300-0632.4034","DOIUrl":"https://doi.org/10.55730/1300-0632.4034","url":null,"abstract":"Millions of people throughout the world suffer from the complicated and crippling condition of chronic pain. It can be brought on by several underlying disorders or injuries and is defined by chronic pain that lasts for a period exceeding three months. To better understand the brain processes behind pain and create prediction models for pain-related outcomes, machine learning is a potent technology that may be applied in Functional magnetic resonance imaging (fMRI) chronic pain research. Data (fMRI and T1-weighted images) from 76 participants has been included (30 chronic pain and 46 healthy controls). The raw data were preprocessed using fMRIprep and then parcellated using five various atlases such as MSDL, Yeo?17, Harvard, Schaefer, and Pauli. Then the functional connectivity between the parcellated Region of Interests (ROIs) has been taken as features for the machine learning classifier models using the Blood Oxygenation Level Dependent (BOLD) signals. To distinguish between those with chronic pain and healthy controls, this study used Support Vector Machines (SVM), Boosting, Bagging, convolutional neural network (CNN), XGboost, and Stochastic Gradient Descent (SDG) classifiers. The classification models use stratified shuffle split sampling to fragment the training and testing dataset during various iterations. Hyperparameter tuning was used to get the best classifier model across several combinations of parameters. The best parameters for the classifier were measured by the accuracy, sensitivity, and specificity of the model. Finally, to identify the top ROIs involved in chronic pain was unveiled by the probability-based feature importance method. The result shows that Pauli (subcortical atlas) and MSDL (cortical atlas) worked well for this chronic pain fMRI data. Boosting algorithm classified chronic pain and healthy controls with 94.35% accuracy on the data parcellated with the Pauli atlas. The top four regions contributing to this classifier model were the extended Amygdala, the Subthalamic nucleus, the Hypothalamus, and the Caudate Nucleus. Also, the fMRI data parcellated using a cortical MSDL atlas was classified using the XGboost model with an accuracy of 87.5%. Left Frontal Pole, Medial Default mode Network, right pars opercularis, dorsal anterior cingulate cortex (dACC), and Front Default mode network are the top five regions that contributed to classify the participants. These findings demonstrate that patterns of brain activity in areas associated with pain processing can be used to categorize individuals as chronic pain patients or healthy controls reliably. These discoveries may help with the identification and management of chronic pain and may pave the way for the creation of more potent tailored medicines for those who suffer from it.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135302712","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
TUĞBERK TAŞ, MUHAMMED ABDULLAH BÜLBÜL, ABAS HAŞİMOĞLU, YAVUZ MERAL, YASİN ÇALIŞKAN, GUNAY BUDAGOVA, MÜCAHİD KUTLU
Dyslexia is a learning disorder, characterized by impairment in the ability to read, spell, and decode letters. It is vital to detect dyslexia in earlier stages to reduce its effects. However, diagnosing dyslexia is a time-consuming and costly process. In this paper, we propose a machine-learning model that predicts whether a Turkish-speaking child has dyslexia using his/her audio records. Therefore, our model can be easily used by smart phones and work as a warning system such that children who are likely to be dyslexic according to our model can seek an examination by experts. In order to train and evaluate, we first create a unique dataset that includes audio recordings of 12 dyslexic children and 13 nondyslexic children in an 8-month period. We explore various machine learning algorithms such as KNN and SVM and use the following features: Mel-frequency cepstral coefficients, reading rate, reading accuracy, the ratio of missing words, and confidence scores of the speech-to-text process. In our experiments, we show that children with dyslexia can be detected with 95.63% accuracy even though we use single-sentence long audio records. In addition, we show that the prediction performance of our model is similar to that of the humans?. In this paper, we provide a preliminary study showing that detecting children with dyslexia based on their audio records is possible. Once the mobile application version of our model is developed, parents can easily check whether their children are likely to be dyslexic or not, and seek professional help accordingly.
{"title":"A machine learning approach for dyslexia detection using Turkish audio records","authors":"TUĞBERK TAŞ, MUHAMMED ABDULLAH BÜLBÜL, ABAS HAŞİMOĞLU, YAVUZ MERAL, YASİN ÇALIŞKAN, GUNAY BUDAGOVA, MÜCAHİD KUTLU","doi":"10.55730/1300-0632.4024","DOIUrl":"https://doi.org/10.55730/1300-0632.4024","url":null,"abstract":"Dyslexia is a learning disorder, characterized by impairment in the ability to read, spell, and decode letters. It is vital to detect dyslexia in earlier stages to reduce its effects. However, diagnosing dyslexia is a time-consuming and costly process. In this paper, we propose a machine-learning model that predicts whether a Turkish-speaking child has dyslexia using his/her audio records. Therefore, our model can be easily used by smart phones and work as a warning system such that children who are likely to be dyslexic according to our model can seek an examination by experts. In order to train and evaluate, we first create a unique dataset that includes audio recordings of 12 dyslexic children and 13 nondyslexic children in an 8-month period. We explore various machine learning algorithms such as KNN and SVM and use the following features: Mel-frequency cepstral coefficients, reading rate, reading accuracy, the ratio of missing words, and confidence scores of the speech-to-text process. In our experiments, we show that children with dyslexia can be detected with 95.63% accuracy even though we use single-sentence long audio records. In addition, we show that the prediction performance of our model is similar to that of the humans?. In this paper, we provide a preliminary study showing that detecting children with dyslexia based on their audio records is possible. Once the mobile application version of our model is developed, parents can easily check whether their children are likely to be dyslexic or not, and seek professional help accordingly.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135246527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Although the standard k-nearest neighbor (KNN) algorithm has been used widely for classification in many different fields, it suffers from various limitations that abate its classification ability, such as being influenced by the distribution of instances, ignoring distances between the test instance and its neighbors during classification, and building a single/weak learner. This paper proposes a novel algorithm, called stepwise dynamic nearest neighbor (SDNN), which can effectively handle these problems. Instead of using a fixed parameter k like KNN, it uses a dynamic neighborhood strategy according to the data distribution and implements a new voting mechanism, called stepwise voting. Experimental results were conducted on 50 benchmark datasets. The results showed that the proposed SDNN method outperformed the KNN method, KNN variants, and the state-of-the-art methods in terms of accuracy.
{"title":"Stepwise dynamic nearest neighbor (SDNN): a new algorithm for classification","authors":"DENİZ KARABAŞ, DERYA BİRANT, PELİN YILDIRIM TAŞER","doi":"10.55730/1300-0632.4016","DOIUrl":"https://doi.org/10.55730/1300-0632.4016","url":null,"abstract":"Although the standard k-nearest neighbor (KNN) algorithm has been used widely for classification in many different fields, it suffers from various limitations that abate its classification ability, such as being influenced by the distribution of instances, ignoring distances between the test instance and its neighbors during classification, and building a single/weak learner. This paper proposes a novel algorithm, called stepwise dynamic nearest neighbor (SDNN), which can effectively handle these problems. Instead of using a fixed parameter k like KNN, it uses a dynamic neighborhood strategy according to the data distribution and implements a new voting mechanism, called stepwise voting. Experimental results were conducted on 50 benchmark datasets. The results showed that the proposed SDNN method outperformed the KNN method, KNN variants, and the state-of-the-art methods in terms of accuracy.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"79 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135246528","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
MEIXIA ZOU, XIUWEN LI, JINZHENG FANG, HONG WEN, WEIWEI FANG
Deep neural networks have recently made remarkable achievements in computer vision applications. However, the high computational requirements needed to achieve accurate inference results can be a significant barrier to deploying DNNs on resource-constrained computing devices, such as those found in the Internet-of-things. In this work, we propose a fresh approach called adaptive channel skipping (ACS) that prioritizes the identification of the most suitable channels for skipping and implements an efficient skipping mechanism during inference. We begin with the development of a new gating network model, ACS-GN, which employs fine-grained channel-wise skipping to enable input-dependent inference and achieve a desirable balance between accuracy and resource consumption. To further enhance the efficiency of channel skipping, we propose a dynamic grouping convolutional computing approach, ACS-DG, which helps to reduce the computational cost of ACS-GN. The results of our experiment indicate that ACS-GN and ACS-DG exhibit superior performance compared to existing gating network designs and convolutional computing mechanisms, respectively. When they are combined, the ACS framework results in a significant reduction of computational expenses and a remarkable improvement in the accuracy of inferences
{"title":"Dynamic deep neural network inference via adaptive channel skipping","authors":"MEIXIA ZOU, XIUWEN LI, JINZHENG FANG, HONG WEN, WEIWEI FANG","doi":"10.55730/1300-0632.4020","DOIUrl":"https://doi.org/10.55730/1300-0632.4020","url":null,"abstract":"Deep neural networks have recently made remarkable achievements in computer vision applications. However, the high computational requirements needed to achieve accurate inference results can be a significant barrier to deploying DNNs on resource-constrained computing devices, such as those found in the Internet-of-things. In this work, we propose a fresh approach called adaptive channel skipping (ACS) that prioritizes the identification of the most suitable channels for skipping and implements an efficient skipping mechanism during inference. We begin with the development of a new gating network model, ACS-GN, which employs fine-grained channel-wise skipping to enable input-dependent inference and achieve a desirable balance between accuracy and resource consumption. To further enhance the efficiency of channel skipping, we propose a dynamic grouping convolutional computing approach, ACS-DG, which helps to reduce the computational cost of ACS-GN. The results of our experiment indicate that ACS-GN and ACS-DG exhibit superior performance compared to existing gating network designs and convolutional computing mechanisms, respectively. When they are combined, the ACS framework results in a significant reduction of computational expenses and a remarkable improvement in the accuracy of inferences","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135246529","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, a spiking neural network (SNN) has been proposed for recognizing the digits written on the LCD screen of an experimental setup. The convergence of the learning algorithm has been ensured by using sliding mode control (SMC) theory and the Lyapunov stability method for the adaptation of the network parameters. The spike response model (SRM) has been utilized in the design of the SNN. The performance of the proposed learning scheme has been evaluated both on the experimental data and on the MNIST dataset. The simulated and experimental results of the SNN structure have been compared with the responses of a conventional neural network (ANN) for which the weight update rules have been also derived using SMC theory. The conducted simulations and experimental studies reveal that convergence can be ensured for the proposed learning scheme and the SNN yields higher recognition accuracy compared to a conventional ANN.
{"title":"Recognizing handwritten digits using spiking neural networks with learning algorithms based on sliding mode control theory","authors":"YEŞİM ÖNİZ, MEHMET AYYILDIZ","doi":"10.55730/1300-0632.4022","DOIUrl":"https://doi.org/10.55730/1300-0632.4022","url":null,"abstract":"In this paper, a spiking neural network (SNN) has been proposed for recognizing the digits written on the LCD screen of an experimental setup. The convergence of the learning algorithm has been ensured by using sliding mode control (SMC) theory and the Lyapunov stability method for the adaptation of the network parameters. The spike response model (SRM) has been utilized in the design of the SNN. The performance of the proposed learning scheme has been evaluated both on the experimental data and on the MNIST dataset. The simulated and experimental results of the SNN structure have been compared with the responses of a conventional neural network (ANN) for which the weight update rules have been also derived using SMC theory. The conducted simulations and experimental studies reveal that convergence can be ensured for the proposed learning scheme and the SNN yields higher recognition accuracy compared to a conventional ANN.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135246666","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this work, the use of t-SNE is proposed to embed 3D point clouds of plants into 2D space for plant characterization. It is demonstrated that t-SNE operates as a practical tool to flatten and visualize a complete 3D plant model in 2D space. The perplexity parameter of t-SNE allows 2D rendering of plant structures at various organizational levels. Aside from the promise of serving as a visualization tool for plant scientists, t-SNE also provides a gateway for processing 3D point clouds of plants using their embedded counterparts in 2D. In this paper, simple methods were proposed to perform semantic segmentation and instance segmentation via grouping the embedded 2D points. The evaluation of these methods on a public 3D plant data set conveys the potential of t-SNE for enabling 2D implementation of various steps involved in automatic 3D phenotyping pipelines.
{"title":"Using t-distributed stochastic neighbor embedding for visualization and segmentation of 3D point clouds of plants","authors":"HELİN DUTAĞACI","doi":"10.55730/1300-0632.4018","DOIUrl":"https://doi.org/10.55730/1300-0632.4018","url":null,"abstract":"In this work, the use of t-SNE is proposed to embed 3D point clouds of plants into 2D space for plant characterization. It is demonstrated that t-SNE operates as a practical tool to flatten and visualize a complete 3D plant model in 2D space. The perplexity parameter of t-SNE allows 2D rendering of plant structures at various organizational levels. Aside from the promise of serving as a visualization tool for plant scientists, t-SNE also provides a gateway for processing 3D point clouds of plants using their embedded counterparts in 2D. In this paper, simple methods were proposed to perform semantic segmentation and instance segmentation via grouping the embedded 2D points. The evaluation of these methods on a public 3D plant data set conveys the potential of t-SNE for enabling 2D implementation of various steps involved in automatic 3D phenotyping pipelines.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135246521","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
VEDAT DELICAN, BEHÇET UĞUR TÖREYİN, EGE ÇETİN, AYLİN YALÇIN SARIBEY
Fingerprints are one of the most important scientific proof instruments in solving forensic cases. Identification in fingerprints consists of three levels based on the flow direction of the papillary lines at the first level, the minutiae points at the second level, and the pores at the third level. The inadequacy of existing imaging systems in detecting fingerprints and the lack of pore details at the desired level limit the widespread use of third-level identification. The fact that fingerprints with images based on pores in the unsolved database are not subjected to any evaluation criteria and remain in the database reveals the importance of the study to be carried out. In this study, different from classical fingerprint identification methods, a direct pore-based identification system for fingerprint matching is proposed with the dataset created by using the Docucenter Nirvis device and Projectina Image Acquisition-7000 software as a hyperspectral imaging system where pores were visualized more clearly. Although difficult from an operational perspective, the pores in the 800 fingerprints in the database were manually marked for the accuracy of the results. Next, by using a scoring based on iterative closest point algorithm, latent fingerprints were found. Results suggest that the higher the number of pores examined and the more accurately the pores were marked, the higher the hit score. At the same time, query results showed that the scores of other sequential fingerprints in the database which came after the matching fingerprint were even lower.
{"title":"Direct pore-based identification for fingerprint matching process","authors":"VEDAT DELICAN, BEHÇET UĞUR TÖREYİN, EGE ÇETİN, AYLİN YALÇIN SARIBEY","doi":"10.55730/1300-0632.4019","DOIUrl":"https://doi.org/10.55730/1300-0632.4019","url":null,"abstract":"Fingerprints are one of the most important scientific proof instruments in solving forensic cases. Identification in fingerprints consists of three levels based on the flow direction of the papillary lines at the first level, the minutiae points at the second level, and the pores at the third level. The inadequacy of existing imaging systems in detecting fingerprints and the lack of pore details at the desired level limit the widespread use of third-level identification. The fact that fingerprints with images based on pores in the unsolved database are not subjected to any evaluation criteria and remain in the database reveals the importance of the study to be carried out. In this study, different from classical fingerprint identification methods, a direct pore-based identification system for fingerprint matching is proposed with the dataset created by using the Docucenter Nirvis device and Projectina Image Acquisition-7000 software as a hyperspectral imaging system where pores were visualized more clearly. Although difficult from an operational perspective, the pores in the 800 fingerprints in the database were manually marked for the accuracy of the results. Next, by using a scoring based on iterative closest point algorithm, latent fingerprints were found. Results suggest that the higher the number of pores examined and the more accurately the pores were marked, the higher the hit score. At the same time, query results showed that the scores of other sequential fingerprints in the database which came after the matching fingerprint were even lower.","PeriodicalId":49410,"journal":{"name":"Turkish Journal of Electrical Engineering and Computer Sciences","volume":"2015 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135246526","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：