Labeled Data Set Research Articles

A new Classifier Chain method of BERT Models For Multi-label Classification of Arabic Abusive Language on Social Media

Moving beyond classical single-label classification tasks proves to be crucial nowadays since the majority of real-life problems are inherently multi-label. Various tasks require involving, for one single instance as an input, several classes simultaneously or none of them. Particularly, detecting abusive texts in Arabic social networks intuitively carries several manifestations with regard to the targets’ referred characteristics like gender, race, sexual orientation, etc. Existing multi-label algorithms either invest considerable complexity to handle labels’ correlation or explicitly ignore this information to deal with each label as an independent binary problem. To remedy the abovementioned problems, this paper proposes a new multi-label classification solution based on deep neural networks. The main idea is to exploit the correlation between all the dataset's labels to break it into ordered subsets. Then, follow that order to implement a chain of BERT classifiers. Each classifier is therefore trained on a corresponding subset and passes its output via feature space to the following one. For a new instance, we gather all the classifiers’ results and pick the best one for each class. Experiments showed that our approach significantly outperformed the single classifier's method using BERT as well as other neural network architectures on our Arabic abusive language dataset. 91%,76%, and 82% were achieved respectively for precision, recall, and F1-score. Moreover, we further investigated the performance of our approach on Kaggle multi-label research paper dataset and achieved very promising results.

Semi-Supervised Machine Learning for Livestock Threat Classification Using GPS Data

South African livestock farmers face major challenges in the form of livestock theft and predation. In response to these concerns, farmers started using a collar that monitors the acceleration of an animal and, when specific parameters are met, triggers an alarm which transmits GPS data to the user’s mobile application. Typically, a collar is placed on one animal per flock or herd. In this work, we aim to classify the GPS trajectories captured by these devices into four categories: theft, predation, own-handling and other. We lay particular emphasis on distinguishing theft alarms since these have direct implications for the safety and financial sustainability of farmers. To date, just over one million of these alarms have been recorded. Unfortunately, these trajectories are not labelled with the four categories. Therefore, we start by collecting labelled data sets that can be used for training classification models. We then investigate supervised and semi-supervised approaches for classifying the trajectories. Our semi-supervised approach shows the best results with performance comparable to human performance. The approach consists of three parts. First, an autoencoder and classifier are jointly trained to produce fixed-dimensional embeddings from GPS trajectories. Second, these embeddings are clustered to produce cluster labels. And lastly, the cluster labels are added to human-engineered features and used to train a final classifier. Our semi-supervised approach achieves an overall classification accuracy of 69%, with an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> score of 56% for theft events (4% lower than human performance) and an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> score of 90% for own-handling events (slightly outperforming a human). This model can be deployed to aid farmers in terms of safety and security by providing them with critical information in emergency situations.

SDSC-UNet: Dual Skip Connection ViT-Based U-Shaped Model for Building Extraction

Benefiting from effective global information interaction, vision-transformers (ViTs) have been widely used in the building extraction task. However, buildings in remote sensing (RS) images usually differ greatly in size. Mainstream ViT-based segmentation models for RS images are based on Swin Transformer, which lacks multi-scale information inside the ViT block. In addition, they only connect the output of the entire ViT encoder block to the decoder, which ignore the similarity information of the attention maps inside the ViT encoder block, and are unable to provide better global dependencies for the decoder. To solve above problems, we introduce a novel Shunted Transformer, which enables the model to capture multi-scale information internally while fully establishing global dependencies, to build a pure ViT-based U-shaped model for building extraction. Furthermore, unlike the previous single-skip-connection structure of U-shaped methods, we build a novel dual skip connection structure inside the model. It simultaneously transmits the attention maps inside the ViT encoder block and its entire output to the decoder, thereby fully mining the information of the ViT encoder block and providing better global information guidance for the decoder. Thus, our model is named Shunted Dual Skip Connection UNet (SDSC-UNet). We also design a feature fusion module called Dual Skip Upsample Fusion Module (DSUFM) to aggregate the information. Our model has yields state-of-the-art (SOTA) performance (83.02%IoU) on the Inria Aerial Image Labeling Dataset. Code will be available.

3D Visualization of Pulmonary Vessel Based on Low-Cost Segmentation and Fast Reconstruction

Real-time visual-aided navigation and path strategy for pneumonoconiosis and efficient 3D visualization of pulmonary vessels are of great research and clinical significance in the treatment of lung diseases. The complex structure of lung tissue limits the application of deep learning in pulmonary vascular visualization due to the lack of vascular labeling datasets. Also, the existing methods have large computational complexity and are low efficiency. This study proposes a method for high-quality 3D visualization of pulmonary vessels based on low-cost segmentation and fast reconstruction, consisting of three steps: 1) Pulmonary vessel feature extraction from lung CT images using self-supervised learning, 2) Segmentation of pulmonary sparse vessels in lung CT images using self-supervised transfer learning, and 3) 3D reconstruction of pulmonary vessels based on segmentation results of step (2) using interpolation. The accuracy of pulmonary vascular contour segmentation was improved from 91.31% using the sparse coding to 98.65% using our proposed method (27,270 test sample points); the classifier evaluation accuracy was improved from 95.33% to 98.26%, and the average running time of the model with the test set data was 44 ms per slice. the segmentation results can automatically generate a complete vascular tree model with an average time of 10.8s ± 1 1.6s. The results demonstrate that the proposed method provides fast and accurate 3D visualization of pulmonary vessels, and is promising for more precise and reliable information for pneumonoconiosis patients.

Hidden Feature-Guided Semantic Segmentation Network for Remote Sensing Images

For semantic segmentation of remote sensing images, convolutional neural networks (CNNs) have proven to be powerful tools. However, the existing CNNs-based methods have the problems of feature information loss, serious interference by clutter information, and ignoring the correlation between different scale features. To solve these problems, this article proposes a novel hidden feature-guided semantic segmentation network (HFGNet) for remote sensing images, which achieves accurate semantic segmentation by hierarchically extracting and fusing valuable feature information. Specifically, the hidden feature extraction module (HFE-M) is introduced to suppress the salient feature representation to mine more valuable hidden features. Meanwhile, the multi-feature interactive fusion module (MIF-M) establishes the correlation between different features to achieve hierarchical feature fusion. The multi-scale feature calibration module (MSFC) is constructed to enhance the diversity and refinement representation of hierarchical fusion features. Besides, the local-channel attention mechanism (LCA-M) is designed to improve the feature perception capability of the object region and suppress background information interference. We conducted extensive experiments on the widely-used ISPRS 2-D Semantic Labeling dataset and the 15-Class Gaofen Image dataset. Experimental results demonstrate that the proposed HFGNet has advantages over several state-of-the-art methods. The source code and models are available at https://github.com/darkseid-arch/RS-HFGNet.

Noise-Sensitive Adversarial Learning for Weakly Supervised Salient Object Detection

Weakly supervised salient object detection (WSOD) aims at training saliency detection models with weak supervision. Normally, the WSOD methods use pseudo labels converted from image-level classification labels to train the saliency network. However, the converted pseudo labels always contain noise information compared to ground truth. Previous methods are directly affected by pseudo label noise to generate error-prone predictions. To mitigate this problem, we design a noise-robust adversarial learning framework and propose a noise-sensitive training strategy for the framework. The framework consists of a saliency network and a noise-robust discriminator network. With the guidance of noise-robust discriminator network, our saliency network is robust to noise information in pseudo labels. The proposed noise-sensitive training strategy can make good use of both superior and inferior samples in the pseudo label dataset. With the noise-sensitive training strategy, our framework can further balance the learning of saliency information and the robustness of noise information. Comprehensive experiments on five public datasets demonstrate that our method outperforms the existing image-level classification label based WSOD methods.

NRN-RSSEG: A Deep Neural Network Model for Combating Label Noise in Semantic Segmentation of Remote Sensing Images

The performance of deep neural networks depends on the accuracy of labeled samples, as they usually contain label noise. This study examines the semantic segmentation of remote sensing images that include label noise and proposes an anti-label-noise network framework, termed Labeled Noise Robust Network in Remote Sensing Image Semantic Segmentation (NRN-RSSEG), to combat label noise. The algorithm combines three main components: network, attention mechanism, and a noise-robust loss function. Three different noise rates (containing both symmetric and asymmetric noise) were simulated to test the noise resistance of the network. Validation was performed in the Vaihingen region of the ISPRS Vaihingen 2D semantic labeling dataset, and the performance of the network was evaluated by comparing the NRN-RSSEG with the original U-Net model. The results show that NRN-RSSEG maintains a high accuracy on both clean and noisy datasets. Specifically, NRN-RSSEG outperforms UNET in terms of PA, MPA, Kappa, Mean_F1, and FWIoU in the presence of noisy datasets, and as the noise rate increases, each performance of UNET shows a decreasing trend while the performance of NRN-RSSEG decreases slowly and some performances show an increasing trend. At a noise rate of 0.5, the PA (−6.14%), MPA (−4.27%) Kappa (−8.55%), Mean_F1 (−5.11%), and FWIOU (−9.75%) of UNET degrade faster; while the PA (−2.51%), Kappa (−3.33%), and FWIoU of NRN-RSSEG (−3.26) degraded more slowly, MPA (+1.41) and Mean_F1 (+2.69%) showed an increasing trend. Furthermore, comparing the proposed model with the baseline method, the results demonstrate that the proposed NRN-RSSEG anti-noise framework can effectively help the current segmentation model to overcome the adverse effects of noisy label training.

Plasma image classification using cosine similarity constrained convolutional neural network

Plasma jets are widely investigated both in the laboratory and in nature. Astrophysical objects such as black holes, active galactic nuclei and young stellar objects commonly emit plasma jets in various forms. With the availability of data from plasma jet experiments resembling astrophysical plasma jets, classification of such data would potentially aid in not only investigating the underlying physics of the experiments but also the study of astrophysical jets. In this work we use deep learning to process all of the laboratory plasma images from the Caltech Spheromak Experiment spanning two decades. We found that cosine similarity can aid in feature selection, classify images through comparison of feature vector direction and be used as a loss function for the training of AlexNet for plasma image classification. We also develop a simple vector direction comparison algorithm for binary and multi-class classification. Using our algorithm we demonstrate 93 % accurate binary classification to distinguish unstable columns from stable columns and 92 % accurate five-way classification of a small, labelled data set which includes three classes corresponding to varying levels of kink instability.

Evaluation of Fetal Head Circumference (HC) and Biparietal Diameter (BPD (Biparietal Diameter)) in Ultrasound Images Using Multi-Task Deep Convolutional Neural Network

Introduction: Ultrasound imaging is a standard examination during pregnancy that can measure specific biometric parameters towards prenatal diagnosis and estimating gestational age. Fetal head circumference (HC) is a significant factor in determining fetus growth and health. Methods: This paper proposes a multi-task deep convolutional neural network for automatic segmentation and estimation of HC (Fetal head circumference) ellipse by minimizing a compound cost function composed of segmentation dice score and MSE of ellipse parameters. Ultrasound-based fetal biometric measurements, such as head circumference (HC) and biparietal diameter (BPD (BIPARIETAL DIAMETER)), are commonly used to evaluate the gestational age and diagnose fetal central nervous system (CNS) pathology. Since manual measurements are operator-dependent and time-consuming, there have been numerous researches on automated methods. However, existing computerized methods still are not satisfactory in terms of accuracy and reliability, owing to difficulties in dealing with various artefacts in ultrasound images. Results: This paper focuses on fetal head biometry and develops a deep-learning-based method for estimating HC (Fetal head circumference) and BPD (BIPARIETAL DIAMETER) with a high degree of accuracy and reliability. Conclusion: The proposed method effectively identifies the head boundary by differentiating tissue image patterns concerning the ultrasound propagation direction. The proposed method was trained with 102 labelled data set and tested to 70 ultrasound images. We achieved a success rate of 92.31% for HC (Fetal head circumference) and BPD (BIPARIETAL DIAMETER) estimations and an accuracy of 87.14% for the plane acceptance check.

Predicting groundwater potential assessment in water-deficient islands based on convolutional neural networks

Due to the special hydrogeological features of the sea islands, the distribution of groundwater is difficult to assess. In order to improve the accuracy of the predicting groundwater potential assessment(GPA) in islands region. We tried to use the convolutional neural networks(CNN) and took lithology, aspect, slope, water density, vegetation fraction, soil humidity and land surface temperature as the remote sensing assessment indicators. The island area of Pearl River Estuary, China was selected as the study area. The groundwater situation of Wailingding Island was taken as the sample data of the total area, and the map of this island was divided into 24 × 22 grids. The levels of GPA of each small grid was determined by hydrogeological maps with results of geophysical method and wells or springs data. Meanwhile, the corresponding data were enlarged and the sample label data set was made. Using the CNN model of learning, training and testing, analysis of groundwater potential and each coupling correlation between remote sensing indicators constantly. After 1500 times of training, loss of model dropped to 0.3113, the accuracy of model was 96.96%. A good 5 levels classification prediction of GPA model was received. The AUC of ROC curve and significance level (P) of the CNN model were 0.855 and 0.001 respectively, which were better than the results of the classic GRSFAI model. The results showed that the prediction of GPA based on CNN model can effectively assess the groundwater distribution levels in the Pearl River Estuary island area, which can provide a certain reference value or a GPA model for the other water-deficient islands. Moreover, the obtained sample set of groundwater volume distribution of bedrock islands can be used as a valuable data source for further in-depth research.

Conditional Fragment Ion Probabilities Improve Database Searching for Nonmonoisotopic Precursors.

Stochastic, intensity-based precursor isolation can result in isotopically enriched fragment ions. This problem is exacerbated for large peptides and stable isotope labeling experiments using deuterium or 15N. For stable isotope labeling experiments, incomplete and ubiquitous labeling strategies result in the isolation of peptide ions composed of many distinct structural isomers. Unfortunately, existing proteomics search algorithms do not account for this variability in isotopic incorporation, and thus often yield poor peptide and protein identification rates. We sought to resolve this shortcoming by deriving the expected isotopic distributions of each fragment ion and incorporating them into the theoretical mass spectra used for peptide-spectrum-matching. We adapted the Comet search platform to integrate a modified spectral prediction algorithm we term Conditional fragment Ion Distribution Search (CIDS). Comet-CIDS uses a traditional database searching strategy, but for each candidate peptide we compute the isotopic distribution of each fragment to better match the observed m/z distributions. Evaluating previously generated D2O and 15N labeled data sets, we found that Comet-CIDS identified more confident peptide spectral matches and higher protein sequence coverage compared to traditional theoretical spectra generation, with the magnitude of improvement largely determined by the amount of labeling in the sample.

Underwater Acoustic Target Recognition Based on Attention Residual Network.

Underwater acoustic target recognition is very complex due to the lack of labeled data sets, the complexity of the marine environment, and the interference of background noise. In order to enhance it, we propose an attention-based residual network recognition method (AResnet). The method can be used to identify ship-radiated noise in different environments. Firstly, a residual network is used to extract the deep abstract features of three-dimensional fusion features, and then a channel attention module is used to enhance different channels. Finally, the features are classified by the joint supervision of cross-entropy and central loss functions. At the same time, for the recognition of ship-radiated noise in other environments, we use the pre-training network AResnet to extract the deep acoustic features and apply the network structure to underwater acoustic target recognition after fine-tuning. The two sets of ship radiation noise datasets are verified, the DeepShip dataset is trained and verified, and the average recognition accuracy is 99%. Then, the trained AResnet structure is fine-tuned and applied to the ShipsEar dataset. The average recognition accuracy is 98%, which is better than the comparison method.

Metric-based rule optimizing system for code smell detection using Salp Swarm and Cockroach Swarm algorithm

Software developers find it difficult to select the specific detection rules for different smell types. A set of metrics, thresholds and labels constitutes a code smells detection rule. The generated rules must be optimized efficiently to ensure successful rule selection. The objective is to identify how rules are generated from the labeled data set and selected using bio-inspired algorithms. The goals are met by employing the C4.5 and RIPPER algorithms to generate rules then, optimized using two bio-inspired algorithms, the Salp Swarm Algorithm (SSA) and Cockroach Swarm Optimization (CSO). The optimized sets of rules are evaluated using the similarity metrics which are computed with the help of expected and the detected code smells. The common rule subsets from SSA and CSO are merged to produce the optimal rule subset which can be used for code smell detection. The proposed work has been experimented on Xerces-J, Log4J, Gantt Project and JFreeChart dataset. The work detected code smells with an accuracy of 91.7% for Xerces-J, 96.7% for JFreeChart, 88.6% for Gantt Project and 98% for Log4J. The findings will be useful for both theory and research since the proposed framework allows focusing on rule selection.

Detecting Cheating in Large-Scale Assessment: The Transfer of Detectors to New Tests.

Recent approaches to the detection of cheaters in tests employ detectors from the field of machine learning. Detectors based on supervised learning algorithms achieve high accuracy but require labeled data sets with identified cheaters for training. Labeled data sets are usually not available at an early stage of the assessment period. In this article, we discuss the approach of adapting a detector that was trained previously with a labeled training data set to a new unlabeled data set. The training and the new data set may contain data from different tests. The adaptation of detectors to new data or tasks is denominated as transfer learning in the field of machine learning. We first discuss the conditions under which a detector of cheating can be transferred. We then investigate whether the conditions are met in a real data set. We finally evaluate the benefits of transferring a detector of cheating. We find that a transferred detector has higher accuracy than an unsupervised detector of cheating. A naive transfer that consists of a simple reuse of the detector increases the accuracy considerably. A transfer via a self-labeling (SETRED) algorithm increases the accuracy slightly more than the naive transfer. The findings suggest that the detection of cheating might be improved by using existing detectors of cheating at an early stage of an assessment period.

Top-k Partial Label Machine.

To deal with ambiguities in partial label learning (PLL), the existing PLL methods implement disambiguations, by either identifying the ground-truth label or averaging the candidate labels. However, these methods can be easily misled by the false-positive labels in the candidate label set. We find that these ambiguities often originate from the noise caused by highly correlated or overlapping candidate labels, which leads to the difficulty in identifying the ground-truth label on the first attempt. To give the trained models more tolerance, we first propose the top-k partial loss and convex top-k partial hinge loss. Based on the losses, we present a novel top-k partial label machine (TPLM) for partial label classification. An efficient optimization algorithm is proposed based on accelerated proximal stochastic dual coordinate ascent (Prox-SDCA) and linear programming (LP). Moreover, we present a theoretical analysis of the generalization error for TPLM. Comprehensive experiments on both controlled UCI datasets and real-world partial label datasets demonstrate that the proposed method is superior to the state-of-the-art approaches.

Exploration of Semantic Label Decomposition and Dataset Size in Semantic Indoor Scenes Synthesis via Optimized Residual Generative Adversarial Networks.

In this paper, we revisit the paired image-to-image translation using the conditional generative adversarial network, the so-called "Pix2Pix", and propose efficient optimization techniques for the architecture and the training method to maximize the architecture's performance to boost the realism of the generated images. We propose a generative adversarial network-based technique to create new artificial indoor scenes using a user-defined semantic segmentation map as an input to define the location, shape, and category of each object in the scene, exactly similar to Pix2Pix. We train different residual connections-based architectures of the generator and discriminator on the NYU depth-v2 dataset and a selected indoor subset from the ADE20K dataset, showing that the proposed models have fewer parameters, less computational complexity, and can generate better quality images than the state of the art methods following the same technique to generate realistic indoor images. We also prove that using extra specific labels and more training samples increases the quality of the generated images; however, the proposed residual connections-based models can learn better from small datasets (i.e., NYU depth-v2) and can improve the realism of the generated images in training on bigger datasets (i.e., ADE20K indoor subset) in comparison to Pix2Pix. The proposed method achieves an LPIPS value of 0.505 and an FID value of 81.067, generating better quality images than that produced by Pix2Pix and other recent paired Image-to-image translation methods and outperforming them in terms of LPIPS and FID.

Eliminate Time Dispersion of Seismic Wavefield Simulation with Semi-Supervised Deep Learning

Finite-difference methods are the most widely used methods for seismic wavefield simulation. However, numerical dispersion is the main issue hindering accurate simulation. In the case where the finite-difference scheme is known, the time dispersion can be predicted mathematically and, thus, can be eliminated. However, when only pre-compiled software is available for wavefield simulation, which is common in practical applications, the software-used algorithm becomes a black box (unknown). Therefore, it is challenging to obtain the mathematical expression of the time dispersion, resulting in difficulty in eliminating the time dispersion. To solve this problem, we propose to use deep learning methods to eliminate time dispersion. We design a semi-supervised framework based on convolutional and recurrent neural networks for eliminating time dispersion caused by seismic wave modeling. The framework of our proposed neural network includes two main modules: Inverse Model and Forward Model, both of which have learnable parameters. The Inverse Model is used for eliminating time dispersion while the Forward Model is used for regularizing the training. Particularly, this framework includes two steps: Firstly, using the compiled modeling software to generate two data sets with large and small time steps. Secondly, we train these two modules for transformation between large time-step data (with time dispersion) and small time-step data (without time dispersion) by labeled and unlabeled data sets. In this work, the labeled data set is a paired data set with large time-step data and their corresponding small time-step data; the unlabeled data set is the large time-step data that need time-dispersion elimination. We use the unlabeled data set to guide the network. In this learning framework, re-training is required whenever the modeling algorithms, time interval, or frequency band is changed. Hence, we propose a transfer learning training method to extend from the trained model to another model, which reduces the computational cost caused by re-training. This minor drawback is offset overwhelmingly by the modeling efficiency gain with large time steps in large-scale production. Tests on two models confirm the effectiveness of the proposed method.

Sample2Sample: an improved self-supervised denoising framework for random noise suppression in distributed acoustic sensing vertical seismic profile data

SUMMARYThe performance of supervised deep learning-based denoising methods relies on massive amounts of high-quality training data set with labels. However, data labelling is a time-consuming and tedious process, and the lack of labelled data set has become a major bottleneck affecting the development of supervised deep learning-based denoising methods. In recent years, denoising methods that only use unlabelled noisy data set for training have received more and more attention. Although these methods get rid of the dependence on labels, they usually have some specific requirements on the training data set. For example, the paired training data are required to be multiple noisy observations for each scene or obey a specific noise distribution, etc., which are often very challenging to meet in practical applications. In this study, we propose an improved self-supervised denoising framework based on Noise2Noise that only uses noisy seismic data set for training, and we name it sample2sample. The proposed denoising framework does not require multiple repeated acquisitions of seismic data to obtain multiple independent noisy observations for each scene used for training, and has no specific requirement for the noise distribution prior. Specifically, we introduce a random sampler to generate paired subsamples from some individual noisy seismic data for training. The corresponding elements in the two paired subsamples are adjacent in the original seismic data and approximately meet the training premise of Noise2Noise, that is the paired training data have the same signal. In addition, considering that there are some subtle differences in the signals of the paired subsamples generated by sampling, we also introduce a regularization loss to compensate for this. We conducted a qualitative and quantitative analysis of the denoising performance of the proposed method through further experiments, including synthetic data experiments and field data experiments.

Domain Adaptive Log Anomaly Prediction for Hadoop System

Hadoop provides a powerful platform that allows reliable, scalable, and distributed processing of massive data sets across a cluster of computers. Log data record events taken place in the Hadoop system that helps to understand system activities and diagnose problems. However, system upgrades and updates often change the syntax and patterns of logs, rendering the machine-learning models that were designed for the legacy system ineffective. Retraining the machine-learning models with new data sets from scratch might improve the accuracy of the machine-learning model. Nevertheless, annotating new data sets is often time consuming and labor intensive. In this article, we propose a domain adaptive log anomaly prediction framework called LogAT to effectively transfer learned knowledge from the existing labeled data set (source domain) to the new unlabeled data set (target domain) by adopting an unsupervised domain adaption method. Furthermore, a hierarchical anomaly knowledge graph has been constructed to represent the domain knowledge that facilitates the subsequent detection and diagnosis of system faults. Extensive experiments have been conducted on public and real-world data sets to validate the effectiveness of the proposed framework as well as each module. Our results show that LogAT achieves superior performance over the state-of-the-art methods for predicting log anomalies and acquiring considerable performance improvement in terms of AUC-ROC score on different Hadoop application data sets.

A Residual-Inception U-Net (RIU-Net) Approach and Comparisons with U-Shaped CNN and Transformer Models for Building Segmentation from High-Resolution Satellite Images.

Building segmentation is crucial for applications extending from map production to urban planning. Nowadays, it is still a challenge due to CNNs’ inability to model global context and Transformers’ high memory need. In this study, 10 CNN and Transformer models were generated, and comparisons were realized. Alongside our proposed Residual-Inception U-Net (RIU-Net), U-Net, Residual U-Net, and Attention Residual U-Net, four CNN architectures (Inception, Inception-ResNet, Xception, and MobileNet) were implemented as encoders to U-Net-based models. Lastly, two Transformer-based approaches (Trans U-Net and Swin U-Net) were also used. Massachusetts Buildings Dataset and Inria Aerial Image Labeling Dataset were used for training and evaluation. On Inria dataset, RIU-Net achieved the highest IoU score, F1 score, and test accuracy, with 0.6736, 0.7868, and 92.23%, respectively. On Massachusetts Small dataset, Attention Residual U-Net achieved the highest IoU and F1 scores, with 0.6218 and 0.7606, and Trans U-Net reached the highest test accuracy, with 94.26%. On Massachusetts Large dataset, Residual U-Net accomplished the highest IoU and F1 scores, with 0.6165 and 0.7565, and Attention Residual U-Net attained the highest test accuracy, with 93.81%. The results showed that RIU-Net was significantly successful on Inria dataset. On Massachusetts datasets, Residual U-Net, Attention Residual U-Net, and Trans U-Net provided successful results.