Low Texture Research Articles

Strong Robust LIO System Based on Event Camera Assistance

Abstract. At present, the LiDAR system can achieve high precision and is extensively used for indoor and outdoor mobile positioning and mapping. However, LiDAR systems still face issues in cluttered environments where strong features are absent, leading to a degradation of the LiDAR-based solution. When the carrier movement involves high-speed or prolonged exposure to the mirror wall, it can cause severe degradation issues or even positioning failures in the laser slam system. Event cameras are vision sensors inspired by biology that exhibit strong robustness in high dynamic and low texture environments, potentially leading to better performance in such environments. However, some issues with event cameras remain unresolved. In this paper, a multi-source fusion method based on EVIO, LIO and IMU trajectory layer post-processing method is proposed, which will fully consider the robustness of event camera in high dynamic environment and the high precision advantage of Lidar in conventional environment, and use an algorithm based on normalized uncertainty. The elastic multi-source fusion of event camera and LiDAR is realized and tested in realde environment. Experimental results show that the proposed algorithm can effectively improve the accuracy of event camera and LiDAR. Compared to the current more advanced algorithms, the proposed algorithm effectively addresses the problem of LIO in degraded environments. Additionally, it mitigates the scale inaccuracy and divergence of the EVIO trajectory to some extent. Compared to LIO, the algorithm can reduce the maximum position error by approximately 30% and increase the overall position accuracy by 32%. Additionally, it can significantly constrain the divergence of errors in the Y direction, improving its accuracy by about 75% and 65% compared to the LIO and EVIO algorithms, respectively.

Microsaccadic suppression of peripheral perceptual detection performance as a function of foveated visual image appearance.

Microsaccades are known to be associated with a deficit in perceptual detection performance for brief probe flashes presented in their temporal vicinity. However, it is still not clear how such a deficit might depend on the visual environment across which microsaccades are generated. Here, and motivated by studies demonstrating an interaction between visual background image appearance and perceptual suppression strength associated with large saccades, we probed peripheral perceptual detection performance of human subjects while they generated microsaccades over three different visual backgrounds. Subjects fixated near the center of a low spatial frequency grating, a high spatial frequency grating, or a small white fixation spot over an otherwise gray background. When a computer process detected a microsaccade, it presented a brief peripheral probe flash at one of four locations (over a uniform gray background) and at different times. After collecting full psychometric curves, we found that both perceptual detection thresholds and slopes of psychometric curves were impaired for peripheral flashes in the immediate temporal vicinity of microsaccades, and they recovered with later flash times. Importantly, the threshold elevations, but not the psychometric slope reductions, were stronger for the white fixation spot than for either of the two gratings. Thus, like with larger saccades, microsaccadic suppression strength can show a certain degree of image dependence. However, unlike with larger saccades, stronger microsaccadic suppression did not occur with low spatial frequency textures. This observation might reflect the different spatiotemporal retinal transients associated with the small microsaccades in our study versus larger saccades.

Ranking the Relative Importance of Image Quality Features in CT by Consensus Survey

ObjectiveThis study sought to determine consensus opinions from subspecialty radiologists and imaging physicists on the relative importance of image quality features in CT. MethodsA prospective survey of subspecialty radiologists and medical physicists was conducted to collect consensus opinions on the relative importance of ten image quality features: axial sharpness, blooming, contrast, longitudinal sharpness, low contrast axial sharpness, metal artifact, motion, noise magnitude, noise texture, and streaking. The survey was first sent to subspecialty radiologists in volunteer leadership roles in the American College of Radiology and Radiological Society of North America, thereafter relying on snowball sampling. Surveyed subspecialties were abdominal, cardiac, emergency, musculoskeletal, neuro, pediatric, and thoracic radiology, and medical physics. Individual respondents’ ratings were normalized for calculation of mean normalized ratings and priority rankings for each feature within subspecialties. Also calculated were intraclass correlation coefficients across image quality features within subspecialties, and analysis of variance across subspecialties within each feature. ResultsMost subspecialties had moderate to excellent intraclass agreement. For every radiology subspecialty except musculoskeletal, motion was the most important image quality feature. There was agreement across subspecialties that axial sharpness and contrast are only moderately important. There was disagreement across subspecialties on the relative importance of noise magnitude. Blooming was highly important to cardiac radiologists, and noise texture was highly important to musculoskeletal radiologists. ConclusionImage quality preferences differ based on clinical tasks and challenges in each anatomical radiology subspecialty. CT image analysis and development of quantitative measures of quality and protocol optimization—and related policy initiatives—should be specific to radiology subspecialty.

High-strength and high-plasticity ZK60 magnesium alloy prepared by rapid solidification and high-speed extrusion

The extrusion speed of ZK60 magnesium alloy is usually lower than 5 m/min. The rapidly solidified (RS) ZK60 magnesium alloy extruded at 350 ℃ with die-exit speeds of 10 m/min, 20 m/min and 30 m/min was investigated. The results reveal that after the RS ZK60 alloy was extruded at different die-exit speeds, the average size of dynamically recrystallized grains is 1.19–1.40 μm, the percent of high-angle grain boundaries (HAGBs) become larger with the average orientation degree of 57.83 ° to 54.81 °, the intensity of texture at (0001)<1̅100> is below 3.4, and the Schmid factors of the prismatic and pyramidal planes reach the high value of no lower than 0.42 overall, while no coarse MgZn phase was observed. The yield tensile strength, ultimate tensile strength and elongation of the RS as-extruded ZK60 magnesium alloy change slowly from 321 to 299 MPa, -355–348 MPa, and -23–27 % when the die-exit speed increase from 10 to 30 m/min, respectively. The ultra-fine grains and disappearance of the coarse low melting-point MgZn phase are the main reasons for the high-speed extrusion, high strength, high plasticity and low texture of the RS as-extruded ZK60 magnesium alloy.

Image Registration Algorithm for Stamping Process Monitoring Based on Improved Unsupervised Homography Estimation

Homography estimation is a crucial task in aligning template images with target images in stamping monitoring systems. To enhance the robustness and accuracy of homography estimation against random vibrations and lighting variations in stamping environments, this paper proposes an improved unsupervised homography estimation model. The model takes as input the channel-stacked template and target images and outputs the estimated homography matrix. First, a specialized deformable convolution module and Group Normalization (GN) layer are introduced to expand the receptive field and enhance the model’s ability to learn rotational invariance when processing large, high-resolution images. Next, a multi-scale, multi-stage unsupervised homography estimation network structure is constructed to improve the accuracy of homography estimation by refining the estimation through multiple stages, thereby enhancing the model’s resistance to scale variations. Finally, stamping monitoring image data is incorporated into the training through data fusion, with data augmentation techniques applied to randomly introduce various levels of perturbation, brightness, contrast, and filtering to improve the model’s robustness to complex changes in the stamping environment, making it more suitable for monitoring applications in this specific industrial context. Compared to traditional methods, this approach provides better homography matrix estimation when handling images with low texture, significant lighting variations, or large viewpoint changes. Compared to other deep-learning-based homography estimation methods, it reduces estimation errors and performs better on stamping monitoring images, while also offering broader applicability.

Exploring the toughening mechanisms of PyC interphase in SiCf/SiC composites through molecular dynamics and mesoscale stochastic simulations

The introduction of pyrolytic carbon (PyC) interphase in SiCf/SiC composites significantly improves their toughness, primarily by deflecting matrix cracks, while the underlying toughening mechanisms toward various PyC microstructures remain mysterious due to the challenges in experimentally observing microscopic deformations within PyC. This paper addresses this gap by constructing distinct PyC models based on orientation angle (OA), a key experimental characteristic, and then conducting Mode I loading simulations on SiCf/PyC/SiC systems by molecular dynamics (MD). Results indicate that as OA increases, the deformation behavior of PyC transits from multilayer sliding to delamination, which correlates to a reduction in the fracture energy of SiCf/PyC/SiC systems. Energy dissipation models are established for two microscopic deformation patterns as multilayer sliding and delamination based on homogenization theory, with results demonstrating that the energy dissipation caused by multilayer sliding within PyC surpasses that of delamination. Furthermore, mesoscale stochastic simulations of crack propagation in PyC with different textures are carried out to obtain crack path configurations and corresponding energy dissipations. The outcomes highlight the superior toughening effect of high texture (HT) PyC over medium texture (MT) and low texture (LT) PyC due to longer crack path within HT PyC, aligning well with experimental findings. This study provides valuable insights into cracking through PyC with varying textures, offering a foundation for optimizing PyC coating processes in SiCf/SiC composites to enhance performance.

PSDF: Prior-Driven Neural Implicit Surface Learning for Multi-view Reconstruction.

Surface reconstruction has traditionally relied on the Multi-View Stereo (MVS)-based pipeline, which often suffers from noisy and incomplete geometry. This is due to that although MVS has been proven to be an effective way to recover the geometry of the scenes, especially for locally detailed areas with rich textures, it struggles to deal with areas with low texture and large variations of illumination where the photometric consistency is unreliable. Recently, Neural Implicit Surface Reconstruction (NISR) combines surface rendering and volume rendering techniques and bypasses the MVS as an intermediate step, which has emerged as a promising alternative to overcome the limitations of traditional pipelines. While NISR has shown impressive results on simple scenes, it remains challenging to recover delicate geometry from uncontrolled real-world scenes which is caused by its underconstrained optimization. To this end, the framework PSDF is proposed which resorts to external geometric priors from a pretrained MVS network and internal geometric priors inherent in the NISR model to facilitate high-quality neural implicit surface learning. Specifically, the visibility-aware feature consistency loss and depth prior-assisted sampling based on external geometric priors are introduced. These proposals provide powerfully geometric consistency constraints and aid in locating surface intersection points, thereby significantly improving the accuracy and delicate reconstruction of NISR. Meanwhile, the internal prior-guided importance rendering is presented to enhance the fidelity of the reconstructed surface mesh by mitigating the biased rendering issue in NISR. Extensive experiments on Tanks and Temples datasets show that PSDF achieves state-of-the-art performance on complex uncontrolled scenes.

A bronchoscopic navigation method based on neural radiation fields.

We introduce a novel approach for bronchoscopic navigation that leverages neural radiance fields (NeRF) to passively locate the endoscope solely from bronchoscopic images. This approach aims to overcome the limitations and challenges of current bronchoscopic navigation tools that rely on external infrastructures or require active adjustment of the bronchoscope. To address the challenges, we leverage NeRF for bronchoscopic navigation, enabling passive endoscope localization from bronchoscopic images. We develop a two-stage pipeline: offline training using preoperative data and online passive pose estimation during surgery. To enhance performance, we employ Anderson acceleration and incorporate semantic appearance transfer to deal with the sim-to-real gap between training and inference stages. We assessed the viability of our approach by conducting tests on virtual bronchscopic images and a physical phantom against the SLAM-based methods. The average rotation error in our virtual dataset is about 3.18 and the translation error is around 4.95 mm. On the physical phantom test, the average rotation and translation error are approximately 5.14 and 13.12 mm. Our NeRF-based bronchoscopic navigation method eliminates reliance on external infrastructures and active adjustments, offering promising advancements in bronchoscopic navigation. Experimental validation on simulation and real-world phantom models demonstrates its efficacy in addressing challenges like low texture and challenging lighting conditions.

A deep learning-based cascade algorithm for pancreatic tumor segmentation.

Pancreatic tumors are small in size, diverse in shape, and have low contrast and high texture similarity with surrounding tissue. As a result, the segmentation model is easily confused by complex and changeable background information, leading to inaccurate positioning of small targets and false positives and false negatives. Therefore, we design a cascaded pancreatic tumor segmentation algorithm. In the first stage, we use a general multi-scale U-Net to segment the pancreas, and we exploit a multi-scale segmentation network based on non-local localization and focusing modules to segment pancreatic tumors in the second stage. The non-local localization module learns channel and spatial position information, searches for the approximate area where the pancreatic tumor is located from a global perspective, and obtains the initial segmentation results. The focusing module conducts context exploration based on foreground features (or background features), detects and removes false positive (or false negative) interference, and obtains more accurate segmentation results based on the initial segmentation. In addition, we design a new loss function to alleviate the insensitivity to small targets. Experimental results show that the proposed algorithm can more accurately locate pancreatic tumors of different sizes, and the Dice coefficient outperforms the existing state-of-the-art segmentation model. The code will be available at https://github.com/HeyJGJu/Pancreatic-Tumor-SEG.

Task-dependent contribution to edge-based versus region-based texture perception

Texture segregation studies indicate that some types of textures are processed by edge-based and others by region-based mechanisms. However, studies employing nominally edge-based textures have found evidence for region-based processing mechanisms when the task was to detect rather than segregate the textures. Here we investigate directly whether the nature of the task determines if region-based or edge-based mechanisms are involved in texture perception. Stimuli consisted of randomly positioned Gabor micropattern texture arrays with five types of modulation: orientation modulation, orientation variance modulation, luminance modulation, contrast modulation and contrast variance modulation (CVM). There were four modulation frequencies: 0.1, 0.2, 0.4 and 0.8 cpd. Each modulation type was defined by three types of waveforms: sinewave (SN) with its smooth variations, square-wave (SQ) and cusp-wave (CS) with its sharp texture edges. The CS waveform was constructed by removing a sinewave from an equal amplitude square-wave. Participants performed two tasks: detection in which participants selected which of two stimuli contained the modulation and discrimination in which participants indicated which of two textures had a different modulation orientation. Our results indicate that threshold amplitudes in the detection task followed the ordering SQ < SN < CS across all spatial frequencies, consistent with detection being mediated by the overall energy in the stimulus and hence region based. With the discrimination task at low texture spatial frequencies and with CVM textures at all spatial frequencies the order was CS ≤ SQ with both < SN, consistent with being edge-based. We modeled the data by estimating the spatial frequency of a Difference of Gaussian filter that gave the largest peak amplitude response to the data. We found that the peak amplitude was lower for detection than discrimination across all texture types except for the CVM texture. We conclude that task requirements are critical to whether edges or regions underpin texture processing.

Invisible gas detection: An RGB-thermal cross attention network and a new benchmark

The widespread use of various chemical gases in industrial processes necessitates effective measures to prevent their leakage during transportation and storage, given their high toxicity. Thermal infrared-based computer vision detection techniques provide a straightforward approach to identify gas leakage areas. However, the development of high-quality algorithms has been challenging due to the low texture in thermal images and the lack of open-source datasets. In this paper, we present the RGB-Thermal Cross Attention Network (RT-CAN), which employs an RGB-assisted two-stream network architecture to integrate texture information from RGB images and gas area information from thermal images. Additionally, to facilitate the research of invisible gas detection, we introduce Gas-DB, an extensive open-source gas detection database including about 1.3K well-annotated RGB-thermal images with eight variant collection scenes. Experimental results demonstrate that our method successfully leverages the advantages of both modalities, achieving state-of-the-art (SOTA) performance among RGB-thermal methods, surpassing single-stream SOTA models in terms of accuracy, Intersection of Union (IoU), and F2 metrics by 4.86%, 5.65%, and 4.88%, respectively. The code and data can be found at https://github.com/logic112358/RT-CAN.

A Super-Resolution and 3D Reconstruction Method Based on OmDF Endoscopic Images.

In the field of endoscopic imaging, challenges such as low resolution, complex textures, and blurred edges often degrade the quality of 3D reconstructed models. To address these issues, this study introduces an innovative endoscopic image super-resolution and 3D reconstruction technique named Omni-Directional Focus and Scale Resolution (OmDF-SR). This method integrates an Omnidirectional Self-Attention (OSA) mechanism, an Omnidirectional Scale Aggregation Group (OSAG), a Dual-stream Adaptive Focus Mechanism (DAFM), and a Dynamic Edge Adjustment Framework (DEAF) to enhance the accuracy and efficiency of super-resolution processing. Additionally, it employs Structure from Motion (SfM) and Multi-View Stereo (MVS) technologies to achieve high-precision medical 3D models. Experimental results indicate significant improvements in image processing with a PSNR of 38.2902 dB and an SSIM of 0.9746 at a magnification factor of ×2, and a PSNR of 32.1723 dB and an SSIM of 0.9489 at ×4. Furthermore, the method excels in reconstructing detailed 3D models, enhancing point cloud density, mesh quality, and texture mapping richness, thus providing substantial support for clinical diagnosis and surgical planning.

Neural network-based reversible data hiding for medical image

Pixel prediction is an important issue in the field of reversible data hiding. Neural networks are gradually used to improve the accuracy of pixel prediction owing to their excellent performance. However, current neural network-based pixel predictors are designed for natural images and do not consider the characteristics of medical images. Therefore, in this paper, we propose a dual-branch neural network-based reversible data hiding scheme for medical images. Detailedly, considering the characteristics of medical images, in which complex and smooth regions are more clearly distinguished, we present a clustering method to classify pixels into three classes according to their complexities, and generate masks to assist pixel prediction. Then, in the prediction stage, a dual-branch neural network-based pixel predictor is designed to extract unique and shared features, and a convolutional block attention module is used to optimize the extracted features. Finally, in the embedding stage, considering the characteristics of region of interest (ROI) and region of non-interest (NROI) in medical images, we design a class-based embedding algorithm, which can prioritize embedding data into NROI with low complexity and then sequentially into low texture complexity region and high texture complexity region of ROI. Experimental results show that our scheme can achieve better performance of pixel prediction and data embedding than existing state-of-the-art works.

Underwater additive manufacturing of 10CrNi3MoV high-strength low-alloy steel by wire-feed laser melting deposition: Effect of microstructure evolution on mechanical properties

The underwater 10CrNi3MoV high-strength low-alloy steel thin-walled part with forming quality up to in-air laser deposition quality was successfully manufactured for the first time. The influencing mechanisms of complicated thermal history and water environment on microstructure evolution and mechanical properties of thin-walled part were researched. Due to austenitization and interlocking effect, the microstructure in the bottom, middle and top (below the 56th layer) areas of underwater sample and the middle area of in-air sample were composed of fine grain ferrite and granular bainite. With the increase of building height, the heat accumulation effect of underwater sample enhanced, resulting in small grain size (GS), high dislocation density, strong solution strengthening effect and low maximum texture strength at the bottom area, while the high proportion of high-angle grain boundaries (HAGBs) at the top area. These resulted in the greatest tensile strength (714 MPa) and microhardness (232 HV) at the bottom area and the highest elongation (11%) at the top area. Due to the cooling effect of water environment, the proportion of HAGBs reduced and there were oxides in fracture. As a result, the horizontal elongation of underwater sample was about 90.9% of in-air sample. Additionally, the GS reduced, and the dislocation density and solution strengthening effect increased, so the microhardness of underwater sample was slightly higher than that of in-air sample. However, due to some microscopic defects, the horizontal and vertical tensile strength of underwater sample were about 97.6% and 90.7% of in-air sample.

Tightly coupled visual-inertial fusion with image enhancement for robust positioning

Abstract Traditional vision-based inertial odometry (VIO) suffers from significant visual degradation, which substantially impacts state estimation in challenging lighting environments. Thermal imaging cameras capture images based on the thermal radiation of objects, rendering them impervious to lighting variations. However, integrating thermal infrared data into conventional visual odometry poses challenges due to its low texture, poor contrast, and high noise levels. In this paper, we propose a tightly coupled approach that seamlessly integrates information from visible light cameras, thermal imaging cameras, and inertial measurement units (IMUs). First, we employ adaptive bilateral filtering and Sobel gradient enhancement to smooth infrared images, thereby reducing noise and enhancing edge contrast. Second, we leverage the Sage-Husa adaptive filter in conjunction with iterative Kalman filtering (IEKF) to effectively mitigate the impact of non-Gaussian noise on the system. Finally, we conduct comprehensive evaluations of the proposed system using both open datasets and real-world experiments across four distinct scenarios: normal lighting, low-light conditions, low-light conditions with camera shake, and challenging lighting environments. Comparative analysis reveals that our method outperforms IEKF, yielding a reduction in localization error measured by root mean square error (RMSE) by 58.69%, 57.24%, 60.23%, and 30.87% in these respective scenarios.

Enhancing thermal facial recognition leveraging large datasets and hybrid algorithms

ABSTRACT Face recognition based on thermal image is a crucial aspect of identity verification that has been developed to counter low or no illumination. This paper proposes a novel hybrid algorithm for thermal face recognition to cope with the low resolution and texture blurring of thermal images. The algorithm contains a multi-scale feature fusion module, an attention module, and a joint loss function, which enhances the feature extraction capability, improves the classification accuracy, and has few network parameters. In addition to the innovative approach, a collaborative thermal facial dataset, named CSU-Laval, has been established by combining the 134 ULFMT dataset from Laval University, Canada, with 210 subjects acquired from Central South University, China. This dataset has 344 subjects and contains a rich set of face variables, including expression, angle, glasses-wearing, and time-lapse.

Multi-view depth estimation based on multi-feature aggregation for 3D reconstruction

Robust 3D reconstruction depends critically on the accuracy of depth maps, and multi-view depth estimate is greatly aided by Multi-View Stereo (MVS). Nevertheless, challenges still exist, particularly in accurately estimating depth in regions with low texture and occlusion edges. We introduce MFA-MVSNet, a novel Multi-Feature Aggregation Multi-View Stereo Network that aggregates diverse features in order to address these issues. The Local Feature Extractor (LFE) is incorporated into MFA-MVSNet to adaptively capture local features across various receptive fields. It also incorporates a Multi-feature Fusion Module (MFM) to combine local features with global features extracted by the Global Feature Extractor (GFE). We utilize neural edge information in a CNN-based Depth Refinement Module (DRM) to iteratively filter the depth map to address depth errors in low-textured regions and occlusion edges. To further improve network robustness and training efficiency, we also introduce a Region Perception Loss (RPL) to lessen the impact of both easily matched and mismatched areas. On both the DTU and BlendedMVS datasets, experimental results demonstrate that MFA-MVSNet outperforms recent advanced methods in terms of depth map quality. Additionally, we incorporate MFA-MVSNet into a multi-view 3D reconstruction pipeline, utilizing its depth maps to enhance downstream tasks such as point cloud reconstruction and neural radiance fields (NeRF), leading to improved 3D reconstruction performance.

Microstructure, properties and damage characteristics of designed PyC interphase in Cf/SiC mini-composites

Pyrolytic carbon (PyC) interphase plays a crucial role in improving the mechanical properties of fiber-reinforced ceramic-matrix composites (FRCMCs). In this paper, three groups of PyC interphase were prepared: the first group (G1) with a mid-high texture PyC (MT-HT-PyC) interphase, the second group (G2) with a low/high texture combination PyC (LT/HT-PyC), and the third group (G3) with a low texture PyC (LT-PyC). The microstructure, intrinsic properties and interfacial properties of these three groups PyC interphase were characterized by SEM, TEM, Raman spectroscopy, nanoindentation tests, and fiber push-in tests. The result showed the LT-PyC with more amorphous PyC lattice and lattice defects is able to reduce the indentation modulus and interfacial properties of PyC interphase. The uniaxial tensile tests revealed that the Cf/SiC mini-composites containing a LT-PyC interphase could significantly improve their tensile strength, exhibiting a 34 % enhancement in tensile strength compared to those counterparts without LT-PyC interphase. The experimental result of scanning ion beam etching (SIBN) also showed that the interfacial crack propagates primarily within the LT-PyC interphase, rather than along the interface between the fiber and the interphase.

Microsaccade-inspired event camera for robotics.

Neuromorphic vision sensors or event cameras have made the visual perception of extremely low reaction time possible, opening new avenues for high-dynamic robotics applications. These event cameras' output is dependent on both motion and texture. However, the event camera fails to capture object edges that are parallel to the camera motion. This is a problem intrinsic to the sensor and therefore challenging to solve algorithmically. Human vision deals with perceptual fading using the active mechanism of small involuntary eye movements, the most prominent ones called microsaccades. By moving the eyes constantly and slightly during fixation, microsaccades can substantially maintain texture stability and persistence. Inspired by microsaccades, we designed an event-based perception system capable of simultaneously maintaining low reaction time and stable texture. In this design, a rotating wedge prism was mounted in front of the aperture of an event camera to redirect light and trigger events. The geometrical optics of the rotating wedge prism allows for algorithmic compensation of the additional rotational motion, resulting in a stable texture appearance and high informational output independent of external motion. The hardware device and software solution are integrated into a system, which we call artificial microsaccade-enhanced event camera (AMI-EV). Benchmark comparisons validated the superior data quality of AMI-EV recordings in scenarios where both standard cameras and event cameras fail to deliver. Various real-world experiments demonstrated the potential of the system to facilitate robotics perception both for low-level and high-level vision tasks.

Towards accurate binocular vision of satellites: A cascaded multi-scale pyramid network for stereo matching on satellite imagery

Stereo matching of high-resolution satellite stereo images is a fundamental and challenging task in photogrammetry, geoscience, and remote sensing. Some deep learning-based methods have shown great potential in solving this task recently. However, these methods have not attracted much attention as their accuracy is still unsatisfactory, especially in regions with intractable factors (e.g., occlusions, repetitive patterns, low texture, non-texture, and disparity discontinuities). To tackle these challenging factors and improve disparity accuracy, we propose an end-to-end stereo matching network named Cascaded Multi-Scale Pyramid Network (CMSP-Net). Firstly, to fully utilize the multi-level spatial context information and mitigate the ambiguous matching caused by repetitive patterns, low texture, non-texture, and occlusions, a cost volume pyramid is constructed with multi-scale image features. Then, each cost volume in the pyramid is aggregated with an hourglass structure to exploit the multi-scale context in a single cost volume. Secondly, considering that high-resolution cost volumes contain more detailed information but have small receptive fields, while low-resolution ones have large receptive fields but lack details, we design an Attention-Guided Cost Fusion (AGCF) strategy to fuse pyramid cost volumes, which effectively integrates the advantages of multi-scale cost volumes while mitigating the backfiring caused by semantic inconsistency across different scales. Thirdly, a disparity refinement module is designed to improve overall disparity accuracy further, which incorporates the reconstruction error and edge clues of input satellite stereo images. Extensive experimental results on multiple satellite datasets demonstrate that the proposed method can effectively reduce matching ambiguities and recover sharp disparity discontinuities, achieving clear superiority over existing state-of-the-art methods.