Imaging Conditions Research Articles

Deep‐salt: Complete three‐dimensional salt segmentation from inaccurate migrated subsurface offset gathers using deep learning

AbstractDelimiting salt inclusions from migrated images during the velocity model building flow is a time‐consuming activity that depends on highly human‐curated analysis and is subject to interpretation errors or limitations of the images and methods available. We propose a supervised deep learning based method to include three‐dimensional salt geometries in the velocity models. We compare two convolutional networks – based on the U‐Net architecture – which can process three‐dimensional seismic data. One architecture uses three‐dimensional convolutional kernels, and the other has convolutional long short‐term memory units. Each architecture requires specific preprocessing steps which affects their training and predictive performance. Both proposed architectures use subsurface offset gathers obtained from reverse time migration with an extended imaging condition as input and are trained to predict the salt inclusions. The velocity model used in migration is a reasonable approximation of sediment velocity but without salt inclusions. Thus, the migration model and, consequently, the migrated images are inaccurate due to the absence of the salt inclusion in the model using just the sediment velocity information for the segmentation. A similar salt inclusion methodology was previously validated for two‐dimensional approaches; we extend it to the three‐dimensional case. Our approach relies on subsurface common image gathers to focus the sediments' reflections around the zero offset and spread salt reflections' energy over large subsurface offsets. The results show that both proposed network models can accurately delineate the salt bodies in the test models, but when evaluating the trained networks for the three‐dimensional SEG/EAGE salt model, the architecture with convolutional long short‐term memory units has proven to generalize better.

An Application of 3D Cross-Well Elastic Reverse Time Migration Imaging Based on the Multi-Wave and Multi-Component Technique in Coastal Engineering Exploration

Precise surveys are indispensable in coastal engineering projects. The extensive presence of sand in the coastal area leads to significant attenuation of seismic waves within unsaturated loose sediments. As a result, it becomes challenging for seismic waves to penetrate the weathered zone and reach the desired depth with significant amount of energy. In this study, the application of three-dimensional (3D) cross-well elastic reverse time migration (RTM) imaging based on multi-wave and multi-component techniques in coastal engineering exploration is explored. Accurate decomposition of vector compressional (P) and shear (S) waves is achieved through two wavefield decoupling algorithms without any amplitude and phase distortion. Additionally, compressional wave pressure components are obtained, which facilitates subsequent independent imaging. This study discusses and analyzes the imaging results of four imaging strategies under cross-correlation imaging conditions in RTM imaging. The analysis leads to the conclusion that scalarizing vector wavefields imaging yields superior imaging of P- and S-waves. Furthermore, the imaging results obtained through this approach are of great physical significance. In order to validate the efficacy of this method in 3D geological structure imaging in coastal areas, RTM imaging experiments were performed on two representative models. The results indicate that the proposed 3D elastic wave imaging method effectively generates accurate 3D cross-well imaging of P- and S-waves. This method utilizes the multi-wave and multi-component elastic wave RTM imaging technique to effectively leverage the Earth’s elastic medium without increasing costs. It provides valuable information about the distribution of subsurface rock layers, interfaces, and other structures in coastal engineering projects. Importantly, this can be achieved without resorting to extensive excavation or drilling operations. This method addresses the limitations of current cross-well imaging techniques, thereby providing abundant and accurate geological and geophysical information for the analysis and interpretation of 3D geological structures in coastal engineering projects. It has important theoretical and practical significance in real-world production, as well as for the study of geological structures in coastal engineering.

Estimation of Shooting Part Using a Camera with Depth Sensors and Pose Estimation Method and Automatic Setting of Optimal X-ray Imaging Conditions

Estimation of Shooting Part Using a Camera with Depth Sensors and Pose Estimation Method and Automatic Setting of Optimal X-ray Imaging Conditions

Optimal Design of Sparse Matrix Phased Array Using Simulated Annealing for Volumetric Ultrasonic Imaging with Total Focusing Method.

The total focusing method (TFM) is often considered to be the 'gold standard' for ultrasonic imaging in the field of nondestructive testing. The use of matrix phased arrays as probes allows for high-resolution volumetric TFM imaging. Conventional TFM imaging involves the use of full matrix capture (FMC) for ultrasonic signals acquisition, but in the case of a matrix phased array, this approach is associated with a huge volume of data to be acquired and processed. This severely limits the frame rate of volumetric imaging with 2D probes and necessitates the use of high-end equipment. Thus, the aim of this research was to develop a novel design method for determining the optimal sparse 2D probe configuration for specific conditions of ultrasonic imaging. The developed approach is based on simulated annealing and involves implementing the solution of the sparse matrix phased array layout optimization problem. In order to implement simulated annealing for the aforementioned task, its parameters were set, the acceptance function was introduced, and the approaches were proposed to compute beam directivity diagrams of sparse matrix phased arrays in TFM imaging. Experimental studies have shown that the proposed approach provides high-quality volumetric imaging with a decrease in data volume of up to 84% compared to that obtained using the FMC data acquisition method.

Compact simultaneous label-free autofluorescence multi-harmonic microscopy for user-friendly photodamage-monitored imaging.

Label-free nonlinear optical microscopy has become a powerful tool for biomedical research. However, the possible photodamage risk hinders further clinical applications. To reduce these adverse effects, we constructed a new platform of simultaneous label-free autofluorescence multi-harmonic (SLAM) microscopy, featuring four-channel multimodal imaging, inline photodamage monitoring, and pulse repetition-rate tuning. Using a large-core birefringent photonic crystal fiber for spectral broadening and a prism compressor for pulse pre-chirping, this system allows users to independently adjust pulse width, repetition rate, and energy, which is useful for optimizing imaging conditions towards no/minimal photodamage. It demonstrates label-free multichannel imaging at one excitation pulse per image pixel and thus paves the way for improving the imaging speed by a faster optical scanner with a low risk of nonlinear photodamage. Moreover, the system grants users the flexibility to autonomously fine-tune repetition rate, pulse width, and average power, free from interference, ensuring the discovery of optimal imaging conditions with high SNR and minimal phototoxicity across various applications. The combination of a stable laser source, independently tunable ultrashort pulse, photodamage monitoring features, and a compact design makes this new system a robust, powerful, and user-friendly imaging platform.

Interactive simulation and visualization of point spread functions in single molecule imaging

The point spread function (PSF) is fundamental to any type of microscopy, most importantly so for single-molecule localization techniques, where the exact PSF shape is crucial for precise molecule localization at the nanoscale. Optical aberrations and fixed fluorophore dipoles often result in non-isotropic and distorted PSFs, impairing and biasing conventional fitting approaches. Further, PSF shapes are deliberately modified in PSF engineering approaches for providing improved sensitivity, e.g., for 3D localization or determination of dipole orientation. As this can lead to highly complex PSF shapes, a tool for visualizing expected PSFs would facilitate the interpretation of obtained data and the design of experimental approaches. To this end, we introduce a comprehensive and accessible computer application that allows for the simulation of realistic PSFs based on the full vectorial PSF model. Our tool incorporates a wide range of microscope and fluorophore parameters, including orientationally constrained fluorophores, as well as custom aberrations, transmission and phase masks, thus enabling an accurate representation of various imaging conditions. An additional feature is the simulation of crowded molecular environments with overlapping PSFs. Further, our app directly provides the Cramér–Rao bound for assessing the best achievable localization precision under given conditions. Finally, our software allows for the fitting of custom aberrations directly from experimental data, as well as the generation of a large dataset with randomized simulation parameters, effectively bridging the gap between simulated and experimental scenarios, and enhancing experimental design and result validation.

Coupled image and kernel prior learning for high-generalized super-resolution

Deep convolution neural networks (DCNNs) have demonstrated great success on single image super resolution, where most existing methods aim to construct a deep model in an fully-supervised way using large amount of synthetic image pairs under ideal degradation assumption. The super-resolution performance would be significantly degraded for the low-resolution images captured under uncontrolled imaging conditions with complicated degradation procedures. To handle the above limitations, this work proposes a high-generalized image super-resolution framework by synergistically learning the latent image and the degradation process to achieve the specific priors for an under-studying observation in an unsupervised manner. Specifically, we incorporate dual branches of networks to configure our framework, where one is structured with both convolution and transformer blocks to learn local-to-global prior to generate high-quality image while the other aims to learn the kernel prior with a simple convolution-based architecture. The kernel prior subnet is pre-trained to enhance the stability of the joint optimization for the overall unsupervised learning procedure. With the estimated degradation kernel and target image, we produce the approximated low-resolution version with a convolution-based degradation block to formulate the loss function for the specific learning. Moreover, to pursue visually plausible image generation, we further incorporate a perceptual loss by leveraging a pre-trained discriminator in Gan-based SR model. Extensive experiments on several benchmark datasets have demonstrated the effectiveness of our proposed high-generalized super-resolution model, manifesting superiority over both supervised and unsupervised state-of-the-art SR models.

Prelaunch Spectral Characterization of the Operational Land Imager-2

The Landsat-9 satellite, launched in September 2021, carries the Operational Land Imager-2 (OLI-2) as one of its payloads. This instrument is a clone of the Landsat-8 OLI and its mission is to continue the operational land imaging of the Landsat program. The OLI-2 instrument is not significantly different from OLI though the instrument-level pre-launch spectral characterization process was much improved. The focal plane modules used on OLI-2 were manufactured as spares for OLI and much of the spectral characterization of the components was performed for OLI. However, while the spectral response of the fully assembled OLI was characterized by a double monochromator system, the OLI-2 spectral characterization made use of the Goddard Laser for Absolute Measurement of Radiance (GLAMR). GLAMR is a system of tunable lasers that cover 350–2500 nm which are fiber-coupled to a 30 in integrating sphere permanently monitored by NIST-traceable radiometers. GLAMR allowed the spectral characterization of every detector of the OLI-2 focal plane in nominal imaging conditions. The spectral performance of the OLI-2 was, in general, much better than requirements. The final relative spectral responses (RSRs) represent the best characterization any Landsat instrument spectral response. This paper will cover the results of the spectral characterization from the component-level to the instrument-level of the Landsat-9 OLI-2.

Autonomous convergence of STM control parameters using Bayesian optimization

Scanning tunneling microscopy (STM) is a widely used tool for atomic imaging of novel materials and their surface energetics. However, the optimization of the imaging conditions is a tedious process due to the extremely sensitive tip–surface interaction, thus limiting the throughput efficiency. In this paper, we deploy a machine learning (ML)-based framework to achieve optimal atomically resolved imaging conditions in real time. The experimental workflow leverages the Bayesian optimization (BO) method to rapidly improve the image quality, defined by the peak intensity in the Fourier space. The outcome of the BO prediction is incorporated into the microscope controls, i.e., the current setpoint and the tip bias, to dynamically improve the STM scan conditions. We present strategies to either selectively explore or exploit across the parameter space. As a result, suitable policies are developed for autonomous convergence of the control parameters. The ML-based framework serves as a general workflow methodology across a wide range of materials.

Self-supervised denoising for multimodal structured illumination microscopy enables long-term super-resolution live-cell imaging

Detection noise significantly degrades the quality of structured illumination microscopy (SIM) images, especially under low-light conditions. Although supervised learning based denoising methods have shown prominent advances in eliminating the noise-induced artifacts, the requirement of a large amount of high-quality training data severely limits their applications. Here we developed a pixel-realignment-based self-supervised denoising framework for SIM (PRS-SIM) that trains an SIM image denoiser with only noisy data and substantially removes the reconstruction artifacts. We demonstrated that PRS-SIM generates artifact-free images with 20-fold less fluorescence than ordinary imaging conditions while achieving comparable super-resolution capability to the ground truth (GT). Moreover, we developed an easy-to-use plugin that enables both training and implementation of PRS-SIM for multimodal SIM platforms including 2D/3D and linear/nonlinear SIM. With PRS-SIM, we achieved long-term super-resolution live-cell imaging of various vulnerable bioprocesses, revealing the clustered distribution of Clathrin-coated pits and detailed interaction dynamics of multiple organelles and the cytoskeleton.

An algorithm for automatic setting of imaging conditions in X-ray CT system

An algorithm for automatic setting of imaging conditions in X-ray CT system

YOLOv8 Analysis for Vehicle Classification Under Various Image Conditions

Purpose: The purpose of this research is to detect vehicle types in various image conditions using YOLOv8n, YOLOv8s, and YOLOv8m with augmentation.Methods: This research utilizes the YOLOv8 method on the DAWN dataset. The method involves using pre-trained Convolutional Neural Networks (CNN) to process the images and output the bounding boxes and classes of the detected objects. Additionally, data augmentation applied to improve the model's ability to recognize vehicles from different directions and viewpoints.Result: The mAP values for the test results are as follows: Without data augmentation, YOLOv8n achieved approximately 58%, YOLOv8s scored around 68.5%, and YOLOv8m achieved roughly 68.9%. However, after applying horizontal flip data augmentation, YOLOv8n's mAP increased to about 60.9%, YOLOv8s improved to about 62%, and YOLOv8m excelled with a mAP of about 71.2%. Using horizontal flip data augmentation improves the performance of all three YOLOv8 models. The YOLOv8m model achieves the highest mAP value of 71.2%, indicating its high effectiveness in detecting objects after applying horizontal flip augmentation. Novelty: This research introduces novelty by employing the latest version of YOLO, YOLOv8, and comparing its performance with YOLOv8n, YOLOv8s, and YOLOv8m. The use of data augmentation techniques, such as horizontal flip, to increase data variation is also novel in expanding the dataset and improving the model's ability to recognize objects.

How to track and segment fish without human annotations: a self-supervised deep learning approach

Tracking fish movements and sizes of fish is crucial to understanding their ecology and behaviour. Knowing where fish migrate, how they interact with their environment, and how their size affects their behaviour can help ecologists develop more effective conservation and management strategies to protect fish populations and their habitats. Deep learning is a promising tool to analyse fish ecology from underwater videos. However, training deep neural networks (DNNs) for fish tracking and segmentation requires high-quality labels, which are expensive to obtain. We propose an alternative unsupervised approach that relies on spatial and temporal variations in video data to generate noisy pseudo-ground-truth labels. We train a multi-task DNN using these pseudo-labels. Our framework consists of three stages: (1) an optical flow model generates the pseudo-labels using spatial and temporal consistency between frames, (2) a self-supervised model refines the pseudo-labels incrementally, and (3) a segmentation network uses the refined labels for training. Consequently, we perform extensive experiments to validate our method on three public underwater video datasets and demonstrate its effectiveness for video annotation and segmentation. We also evaluate its robustness to different imaging conditions and discuss its limitations.

Endoscopic Hyperspectral Imaging System to Discriminate Tissue Characteristics in Tissue Phantom and Orthotopic Mouse Pancreatic Tumor Model.

In this study, we developed an endoscopic hyperspectral imaging (eHSI) system and evaluated its performance in analyzing tissues within tissue phantoms and orthotopic mouse pancreatic tumor models. Our custom-built eHSI system incorporated a liquid crystal tunable filter. To assess its tissue discrimination capabilities, we acquired images of tissue phantoms, distinguishing between fat and muscle regions. The system underwent supervised training using labeled samples, and this classification model was then applied to other tissue phantom images for evaluation. In the tissue phantom experiment, the eHSI effectively differentiated muscle from fat and background tissues. The precision scores regarding fat tissue classification were 98.3% for the support vector machine, 97.7% for the neural network, and 96.0% with a light gradient-boosting machine algorithm, respectively. Furthermore, we applied the eHSI system to identify tumors within an orthotopic mouse pancreatic tumor model. The F-score of each pancreatic tumor-bearing model reached 73.1% for the KPC tumor model and 63.1% for the Pan02 tumor models. The refined imaging conditions and optimization of the fine-tuning of classification algorithms enhance the versatility and diagnostic efficacy of eHSI in biomedical applications.

A simple image correlation technique for imaging subsurface damage from low-velocity impacts in composite structures

A robust computer vision system is proposed to visualize subsurface barely visible impact damage (BVID) in composite structures through a simple image correlation technique together with a damage imaging condition. This system uses a digital camera to record a video of the surface motion, capturing micron-scale dynamic movement from guided waves propagating on the surface of the structure generated via a sweeping frequency excitation (chirp) up to the ultrasonic frequency range. As the excitation frequency changes during the chirp, waves become trapped within this damaged region, forming standing waves to generate local resonance at specific frequencies. This localized resonance accumulates high wave energy in the region, generating a higher transverse displacement at the damage site compared to the remaining part of the structure. In this work, a simple image correlation technique is proposed to correlate each filtered video frame with the temporal mean of the filtered wavefield video to highlight standing waves at localized damage. The proposed image correlation technique is distinct from digital image correlation (DIC) since it does not use correlation for subset matching to track the movement of surface patterns between two images (video frames). Instead, it uses correlation to directly quantify similarities between corresponding image pixels (windows). Utilizing a single camera greatly simplifies system complexity and hence enhances the practicality and potential for real-time performance over a recently developed technique that utilized 3D DIC with a stereo camera for vision-based BVID detection. To realize the aim, work was conducted in two sequential steps: (1) off-axis 2D DIC was employed rather than 3D DIC to examine the potential of employing a single camera to capture images and extract not only in-plane displacement but also a fraction of transverse displacement in which local resonance is dominant and (2) image correlation was then employed, supplanting the off-axis 2D DIC image processing involved in the first step, to highlight the damage with significantly less processing complexity. This proposed technique using a zero-mean normalized cross-correlation imaging condition, rather than the total wave energy used for DIC-based approaches, is efficient and effective for identifying regions of minute surface movement from local resonance within the damage region without the use of the computationally intensive interpolation to get the sub-pixel gray level information followed by subset matching employed in DIC-based image processing. Two geometrically identical CFRP composite honeycomb panels that had been subjected to low-velocity impacts were used for verification and validation, and two excitation location configurations were tested for each panel. Damaged images produced with correlation for a 100 mm × 100 mm field of view using a single 3-s video, or a total of 19,200 video frames, show accurate damage imaging capabilities regardless of excitation location that exceed that of DIC techniques and is comparable to benchmark damage images obtained from laser Doppler vibrometry, ultrasonic C-scans, and X-ray CT scans. This success of correlation-based imaging of subsurface BVID demonstrates substantial improvements in efficiency and practicality and shows high potential for in situ and real-time computer vision-based nondestructive inspection or structural health monitoring of subsurface BVID in composite aircraft and other critical structures.

Consistency of the inverse scattering imaging condition, the energy norm imaging condition and the impedance kernel in acoustic and elastic reverse-time migration

Abstract In this work, we draw connections between the imaging conditions using the impedance kernel, the inverse scattering imaging condition, and the energy norm imaging condition in acoustic and elastic reverse-time migration (RTM). Traditional RTM often introduces large low-wavenumber artifacts that degrade image quality in intricate geological structures with large velocity variations. In practice, the Laplacian filter is commonly used to remove these low-wavenumber artifacts, but it changes the image wavenumber spectrum. The advanced imaging conditions of the inverse scattering, the energy norm, and the impedance kernel can effectively remove the low-wavenumber artifacts while not changing the wavenumber spectrum. This study aims to build a connection between these three types of imaging conditions by conducting detailed analysis in the wavenumber domain for acoustic and elastic RTMs. We find that they are exactly the same except for the varying weights of the source-receiver wavefield cross-correlation. All three imaging conditions can generate clear RTM images that are not affected by low-wavenumber artifacts. Numerical examples for a simple model, Sigsbee 2a, and BP models verify the consistency of these three imaging conditions and show their advantage over conventional simple zero-lag cross-correlation imaging conditions. This is important for improving the quality and reliability of seismic imaging technology.

Integrating bi-temporal VHR optical and long-term SAR images for built-up area change detection

ABSTRACT With the rapid expansion of urbanization, it is imperative to monitor built-up areas changes to promote the sustainable development of cities, aligning with the goals of Sustainable Development Goal 11(SDG 11). Remote sensing big data is valuable for automatically mapping these changes. Bi-temporal very high-resolution (Bi-VHR) optical images have been widely utilized for fine-grained change detection (CD). However, the significant spatiotemporal inconsistency due to imaging conditions and seasonal variations poses challenges for VHR optical CD. Unlike optical images, synthetic aperture radar (SAR) images are unaffected by atmospheric interference and provide robust spatiotemporal features as a supplement. Previous CD algorithms with SAR overlooked the exploration of long-term features of time series. In this study, we propose a novel CD framework combining long-term SAR with Bi-VHR images. It incorporates a spatial-frequency learning module to enhance SAR temporal features and a multisource feature fusion module to adaptively fuse the heterogeneous features. The experiments are conducted on OS-BCD dataset, which is the first dataset specifically designed for this task. The results demonstrate that our proposal outperforms advanced CD methods with F1 score, IoU and OA of 64.99%, 48.13%, and 95.11%, validating the efficacy of our proposal in accurately detecting changes in built-up areas.

Target Detection Adapting to Spectral Variability in Multi-Temporal Hyperspectral Images Using Implicit Contrastive Learning

Hyperspectral target detection (HTD) is a crucial aspect of remote sensing applications, aiming to identify targets in hyperspectral images (HSIs) based on their known prior spectral signatures. However, the spectral variability resulting from various imaging conditions in multi-temporal hyperspectral images poses a challenge to both classical and deep learning (DL) methods. To overcome the limitations imposed by spectral variability, an implicit contrastive learning-based target detector (ICLTD) is proposed to exploit in-scene spectra in an unsupervised way. First, only prior spectra are utilized for explicit supervision, while an implicit contrastive learning module (ICLM) is designed to normalize the feature distributions of prior and in-scene spectra. This paper theoretically demonstrates that the ICLM can transfer the gradients from prior spectral features to those of in-scene spectra based on their feature similarities and differences. Because of transferred gradient signals, the ICLTD is regularized to extract similar representations for the prior and in-scene target spectra, while augmenting feature differences between the target and background spectra. Additionally, a local spectral similarity constraint (LSSC) is proposed to enhance the capability of scene adaptation by leveraging the spectral similarities among in-scene targets. To validate the performance of the ICLTD under spectral variability, multi-temporal HSIs captured under various imaging conditions are collected to generate prior spectra and in-scene spectra. Comparative evaluations against several DL detectors and classical methods reveal the superior performance of the ICLTD in achieving a balance between target detectability and background suppressibility under spectral variability.

Phantom study and clinical application of total-body 18F-FDG PET/CT imaging: How to use small voxel imaging better?

BackgroundConventional PET/CT imaging reconstruction is typically performed using voxel size of 3.0–4.0 mm in three axes. It is hypothesized that a smaller voxel sizes could improve the accuracy of small lesion detection. This study aims to explore the advantages and conditions of small voxel imaging on clinical application.MethodsBoth NEMA IQ phantom and 30 patients with an injected dose of 3.7 MBq/kg were scanned using a total-body PET/CT (uEXPLORER). Images were reconstructed using matrices of 192 × 192, 512 × 512, and 1024 × 1024 with scanning duration of 3 min, 5 min, 8 min, and 10 min, respectively.ResultsIn the phantom study, the contrast recovery coefficient reached the maximum in matrix group of 512 × 512, and background variability increased as voxel size decreased. In the clinical study, SUVmax, SD, and TLR increased, while SNR decreased as the voxel size decreased. When the scanning duration increased, SNR increased, while SUVmax, SD, and TLR decreased. The SUVmean was more reluctant to the changes in imaging matrix and scanning duration. The mean subjective scores for all 512 × 512 groups and 1024 × 1024 groups (scanning duration ≥ 8 min) were over three points. One false-positive lesion was found in groups of 512 × 512 with scanning duration of 3 min, 1024 × 1024 with 3 min and 5 min, respectively. Meanwhile, the false-negative lesions found in group of 192 × 192 with duration of 3 min and 5 min, 512 × 512 with 3 min and 1024 × 1024 with 3 min and 5 min were 5, 4, 1, 4, and 1, respectively. The reconstruction time and storage space occupation were significantly increased as the imaging matrix increased.ConclusionsPET/CT imaging with smaller voxel can improve SUVmax and TLR of lesions, which is advantageous for the diagnosis of small or hypometabolic lesions if with sufficient counts. With an 18F-FDG injection dose of 3.7 MBq/kg, uEXPLORER PET/CT imaging using matrix of 512 × 512 with 5 min or 1024 × 1024 with 8 min can meet the image requirements for clinical use.

From lines to Polygons: Polygonal building contour extraction from High-Resolution remote sensing imagery

Automated extraction of polygonal building contours from high-resolution remote sensing images is important for various applications. However, it remains a difficult task to achieve automated extraction of polygonal buildings at the level of human delineation due to diverse building structures and imperfect image conditions. In this paper, we propose Line2Poly, an end-to-end approach that uses feature lines as geometric primitives to achieve polygonal building extraction by recovering topological relationships among these lines within an individual building. To extract building feature lines with precision, we adopt a two-stage strategy that combines Convolutional Neural Network (CNN) and transformer architectures. A CNN-based module extracts preliminary feature lines, which serve as positional priors for initializing positional queries in the subsequent transformer-based module. For polygonal building contour reconstruction, we devise a learnable polygon topology reconstruction module that predicts adjacency relationships among discrete lines, and integrates lines into building polygons. The resultant building polygons, based on feature lines, exhibit inherent regularity that aligns with manual labeling standards. Extensive experiments on the Vectorizing World Buildings dataset, the WHU aerial building dataset and the WHU-Mix (vector) dataset validate Line2Poly’s impressive performance in building feature line extraction and instance-level building detection. Moreover, Line2Poly’s predictions exhibit the highest level of concurrence with manual delineations, with over 83% agreement on the WHU aerial building test set and 68.7/59.7% on the WHU-Mix (vector) test set I and II, respectively.