Search

Three-dimensional (3D) single-object tracking (3D SOT) is a fundamental yet not well-solved problem in 3D vision, where the complexity of feature matching and the sparsity of point clouds pose significant challenges. To handle abrupt changes in appearance features and sparse point clouds, we propose a novel 3D SOT network, dubbed CDTracker. It leverages both cosine similarity and an attention mechanism to enhance the robustness of feature matching. By combining similarity embedding and attention assignment, CDTracker performs template and search area feature matching in a coarse-to-fine manner. Additionally, CDTracker addresses the problem of sparse point clouds, which commonly leads to inaccurate tracking. It incorporates relatively dense sampling based on the concept of point cloud segmentation to retain more target points, leading to improved localization accuracy. Extensive experiments on both the KITTI and Waymo datasets demonstrate clear improvements in CDTracker over its competitors.

Indoor 3D reconstruction is particularly challenging due to complex scene structures involving object occlusion and overlap. This paper presents a hybrid indoor reconstruction method that segments the room point cloud into internal and external components, and then reconstructs the room shape and the indoor objects in different ways. We segment the room point cloud into internal and external points based on the assumption that the room shapes are composed of some large external planar structures. For the external, we seek for an appropriate combination of intersecting faces to obtain a lightweight polygonal surface model. For the internal, we define a set of features extracted from the internal points and train a classification model based on random forests to recognize and separate indoor objects. Then, the corresponding computer aided design (CAD) models are placed in the target positions of the indoor objects, converting the reconstruction into a model fitting problem. Finally, the indoor objects and room shapes are combined to generate a complete 3D indoor model. The effectiveness of this method is evaluated on point clouds from different indoor scenes with an average fitting error of about 0.11 m, and the performance is validated by extensive comparisons with state-of-the-art methods.

Gas field production forecast is an important basis for decision-making in the gas industry. How to accurately predict the dynamic production during gas field development is an important content of reservoir engineering research. Reservoir numerical simulation is the most common method for predicting oil and gas production. However, it requires a lot of data to build an accurate geological model which is tedious and time-consuming. At present, many scholars have used machine learning and data mining methods to predict oil and gas production, but they have not considered whether the use of increasing production measures will affect the predicted results.Thus, ARIMA-RTS optimal smooth algorithm is the first applied to establish the prediction model of gas well production. According to the historical production data, the model is processed, the production differential autoregressive integral moving average (ARIMA) model in time series is established, then ARIMA model is combined with RTS (Rauch Tung Striebel) smoothing, and the production prediction model is constructed. RTS smoothing algorithm is an enhanced version of Kalman filter. The measurements are firstly processed by the forward filter, and then, a separate backward smoothing pass is used for obtaining the smoothing solution. The correctness of ARIMA-RTS model was verified with the actual production data. The results show that the prediction based on ARIMA-RTS model can accurately reflect the production performance of gas well. This method can effectively reduce the error caused by stimulation when predicting. When using the ARIMA-RTS model and the ARIMA-Kalman model to predict the production of the same gas well, the prediction accuracy of ARIMA-RTS model is higher than that of ARIMA-Kalman model in production wells with stimulation. Compared with that of the ARIMA-Kalman model, the mean relative error fitted by the ARIMA-RTS model is reduced by 46.3%, and the relative mean square error is reduced by 56.48%. ARIMA-RTS optimal smooth algorithm improves the prediction accuracy of gas well that uses stimulation. We therefore conclude that the ARIMA-RTS optimal smooth algorithm can help us better forecast the forecasting gas well production with stimulation, as well as other fuels output.

Object detection is one of the most fundamental yet challenging research topics in the domain of computer vision. Recently, the study on this topic in aerial images has made tremendous progress. However, complex background and worse imaging quality are obvious problems in aerial object detection. Most state-of-the-art approaches tend to develop elaborate attention mechanisms for the space-time feature calibrations with arduous computational complexity, while surprisingly ignoring the importance of feature calibrations in channel-wise. In this work, we propose a simple yet effective calibrated-guidance (CG) scheme to enhance channel communications in a feature transformer fashion, which can adaptively determine the calibration weights for each channel based on the global feature affinity correlations. Specifically, for a given set of feature maps, CG first computes the feature similarity between each channel and the remaining channels as the intermediary calibration guidance. Then, rerepresenting each channel by aggregating all the channels weighted together via the guidance operation. Our CG is a general module that can be plugged into any deep neural networks, which is named as CG-Net. To demonstrate its effectiveness and efficiency, extensive experiments are carried out on both oriented object detection task and horizontal object detection task in aerial images. Experimental results on two challenging benchmarks (i.e., DOTA and HRSC2016) demonstrate that our CG-Net can achieve the new state-of-the-art performance in accuracy with a fair computational overhead. The source code has been open sourced at https://github.com/WeiZongqi/CG-Net.

Gas reservoir numerical simulation is an important method to optimize the development strategy of shale gas reservoirs which has been influenced by the multi-stage fracture. The regular fracture network model was used to build a conventional numerical simulation, in which it was difficult to show the true situation of fracture propagation. However, the physical parameters not only affect the production, but also influence the stimulation effect; moreover, the quality of the fracturing effect also affects the production which causes the input and out parameters to be inaccurate. To solve this problem, the process simulation must be completed from geology to engineering to gas reservoir. The main controlling factors of production are identified with geological and engineering factors such as horizontal stage length, the volume of fracturing fluid, well spacing, production allocation, and proppant mass. Therefore, on the basis of the integrated simulation method of a hydraulic fracturing network simulation and an unstructured grid high-precision numerical simulation, this paper builds an integrated numerical simulation of a shale gas reservoir coupled with geology and engineering to optimize the development strategy with production as the target. Taking four wells of a platform as an example, the EUR (estimated ultimate recovery) has increased by 25% after the optimization of the development strategy.

Unstructured point clouds are a representative shape representation of real-world scenes in 3D vision and graphics. Incompletion inevitably arises, due to the way the set of unorganized points is captured, e.g., as fusion of depth images, merged laser scans, or structure-from-x. In this paper, an end-to-end sparse-to-dense multi-encoder neural network (termed an SDME-Net) is proposed for uniformly completing an unstructured point cloud with its shape details preserved. Unlike most existing learning-based shape completion methods that are enforced on the representations of 2D images and 3D voxelization of point clouds, and require priors of the underlying shape's structures, topologies and annotations, the SDME-Net is implemented on the incomplete and even noisy point cloud without any transformation, and makes no specific assumptions about the incompletion distribution and geometry features in the input. Specifically, the defective point cloud is completed and optimized in a sparse-to-dense manner of two-stages. In the first stage, we generate a sparse but complete point cloud based on a bistratal PointNet, and in the second stage, we yield a dense and high-fidelity point cloud by encoding and decoding the sparse result in the first stage using PointNet++. Meanwhile, we combine the distance loss and repulsion loss to generate more uniformly distributed output point clouds closer to the ground-truth counterparts. Qualitative and quantitative experiments on the public ShapeNet dataset illustrate that our approach outperforms the state-of-art learning-based point cloud shape completion methods in terms of real structure recovery, uniformity, and noise/partiality robustness.

Bas-relief, usually attached to a certain plane, is a semi-stereoscopic sculpture that is suitable for decoration of various environments. Since the rise of 3D printing technology, digital bas-relief modeling is greeted with an avalanche of publicity. In this work, for the detail loss caused by depth compression, we propose a bas-relief modeling method based on the idea of feature separation and migration, which can not only reduce the overall height, but also preserve the original details well. Different from the Weighted Least Squares filter with edge-preserving property, the result of which is affected by gradient changes in the input image, we propose the Full Least Squares filter that allows each region of the input image to receive the same level of smoothing. We first separate the features of bas-relief through the Full Least Squares filter. The details and edges are stripped out together and then combined with the deeply compressed structures to generate a flat bas-relief, in which the edge sinking phenomenon is subsequently optimized to ensure a natural transition between the content and the background. More than that, the watershed algorithm is used to segment the height map of bas-relief, and local detail enhancement or local depth constraint is implemented through mask operation to improve the local feature significance of bas-relief. Experiments show that our method effectively generates the bas-relief with strong flatness and well-preserved details. What's more, the global and local adjustments of features produce rich and diverse planarization effects, which provides more choices for diverse designs of digital bas-relief.

Image texture smoothing (ITS) aims at completely removing textures while preserving as much as possible different-scale structures of an image. However, few quality assessment metrics have been formulated to objectively evaluate the ITS results, due to the lack of ground-truth texture-free images. Considering both the smoothness of textures and the preservation of structures, we make the debut of objective evaluation of ITS results, and design an intuitive Texture Smoothness and Structural Similarity Index (T3SI) based on human perception. Specifically, we first intuitively select some patches which contain relatively more texture/structure information. We then employ the Edge Preservation Index and the Structural Similarity Index Measure to evaluate the texture smoothness and the structure similarity on the selected patches, respectively. We finally formulate the novel exponential entropy function to balance the texture smoothness and the structure similarity for the quality assessment of ITS. T3SI is objective, easy to use, simple to implement, and stable. An independent User Study is performed to verify our proposed T3SI, and large experiments show that T3SI competes successfully to the state-of-the-art metrics. Our code is publicly available.

Land cover classification provides updated information regarding the Earth's resources, which is vital for agricultural investigation, urban management, and disaster monitoring. Current advances in sensor technology on satellite and aerial remote sensing (RS) devices have improved the spatial-spectral, radiometric, and temporal resolutions of images over time. These improvements offer invaluable chances of understanding land cover information. However, land cover classification from RS images is an intricate task because of the high intra-class disparities, low inter-class similarities, and image variation types. We propose a cascaded residual dilated network (CRD-Net) for land cover classification using very high spatial resolution (VHSR) images to address these challenges. The proposed hybrid network follows the encoder-decoder concept with a spatial attention block to guide the network on learnable discriminate features coupled with an intermediary loss to enhance the training process. Moreover, a cascaded residual dilated module increases the network's receptive field to enrich multi-contextual features further, thus boosting the resultant feature descriptor. Extensive experimental results demonstrate that the proposed CRD-Net outperformed state-of-the-art methods, achieving an overall accuracy (OA) of 90.73% and 90.51% on the ISPRS Potsdam land cover dataset and ISPRS Vaihingen dataset, respectively.

Over the years, advances in sensor technologies have enhanced spatial, temporal, spectral, and radiometric resolutions, thus significantly improving the size, resolution, and quality of imagery. These vast developments have inspired improvement in various hyperspectral images (HSI) classification applications such as land cover mapping, vegetation classification, urban monitoring, and understanding which are essential for better utilization of Earth’s resources. HSI classification requires superior algorithms with greater accuracy, less computational complexity, and robustness to extract rich, spectral-spatial information. Deep convolution neural networks (DCCNs) have revolutionized image classification experience, with robust architectures being proposed from time to time. However, insufficient training samples have been earmarked as a significant bottleneck for supervised HSI classification and have not been fully explored in literature. To stimulate further research, this paper reviews current methods that handle labeled data insufficiency and the current feature learning methods for HSI classification using DCNNs. It also presents various methods’ results on the three most popular public HSI datasets, together with intuitive observations motivating future research by the hyperspectral community.

Carbonate fracture-vuggy reservoirs are one of the hot spots in oil and gas exploration and development. However, it is extremely difficult to describe the internal spatial structure of the fracture-vuggy unit and understand the interwell connection relationship. As a method to measure reservoir characteristics and feedback reservoir production information directly according to the detected concentration curve, interwell tracer technology provides a direct measure for people to understand the law of oil-water movement and reservoir heterogeneity and is widely used in various domestic oil fields. Based on the flow law of tracer and the CFD flow simulation basic model, this paper establishes the physical conceptual model and studies the influence of three physical parameters (the flow velocity of the fluid passing through the connected channel, diameter of the connected channel, and length of the connected channel) on the concentration curve at the outlet. In addition, the influence of different interwell connection modes on tracer concentration was studied and classified scientifically. According to the simulation, the tracer concentration curve can be classified into three types: unimodal curve, bimodal curve, and multimodal curve. Finally, the injection-production well group in the T-well area of the Tahe Oilfield is taken as an example, the connection mode between injection and production wells in this well area is further discussed and has been verified, which can be used as a reference for the connectivity analysis of similar carbonate reservoirs.

Distributed temperature sensing (DTS) has been used for fracture parameter diagnosis and flow profile monitoring. In this paper, we present a new model for predicting the temperature profile of two-phase flow multistage fractured horizontal wells in the tight oil reservoirs. The homogeneous reservoir flow/heat transfer model is extended to the tight oil reservoir-fracture-wellbore coupled flow/thermal model. The influence of SRV area on reservoir and wellbore is considered, and the Joule-Thomson effect, heat convection, heat conduction, and other parameters are introduced into the improved model. The temperature distributions of reservoir and wellbore with different production times, water cut, and locations of water entry are simulated. The simulated results indicate that the Joule-Thomson effect will cause wellbore temperature to rise; the temperature of fractures with more water production is significantly lower than that of other fractures, and the water outlet location can be judged according to the temperature change of the wellbore. By using the improved temperature prediction model, the DTS monitoring data of two-phase flow multistage fractured horizontal well in the tight reservoir has been calculated and analyzed, and the accurate production profile has been obtained.

In this paper, we propose an end-to-end single-image super-resolution neural network by leveraging hybrid multi-scale features of images. Different from most existing convolutional neural network (CNN) based solutions, our proposed network depends on the observation that image features extracted by CNN contain hybrid multi-scale features: both multi-scale local texture features and global structural features. By effectively exploiting these multi-scale and local-global features, our network involves far fewer parameters, leading to a large decrease in memory usage and computation during inference. Our network benefits from three key modules: (1) an efficient and lightweight feature extraction module (EFblock); (2) a hybrid multi-scale feature enhancement module (HMblock); and (3) a reconstruction–restoration module (DRblock). Experiments on five popular benchmarks demonstrate that our super-resolution approach achieves better performance with fewer parameters and less memory consumption, compared to more than 20 SOTAs. In summary, we propose a novel multi-scale super-resolution neural network (HMSF), which is more lightweight, has fewer parameters, and requires less execution time, but has better performance than the state-of-the-art methods. Compared to SOTAs, this method is more practical and better suited to run on constrained devices, such as PCs and mobile devices, without the need for a high-performance server.

Laser scan registration estimates a relative transformation to match one scan with another, based on the shape of the overlapping portions of the scans. The core and challenging problem of scan registration in a large-scale scene is, how to detect public targets among consecutive scans accurately and efficiently. We propose an automatic approach to detect cross-shaped targets (consisting of low-cost patterned paper) in raw LiDAR data. We exploit both radiometric and geometric information to fix cross-shaped targets, wherein a Morse function is formulated based on the discrete Morse theory on intensity image to detect the unique target pattern. Benefiting from both strategies of the imageand the geometry-based filtering, the target's center can be determined with sub-point precision. A detailed quantitative and qualitative comparison of the proposed algorithm with the state-of-the-art methods is presented. Results with real data from various complicated scenes, some of which have been successfully applied in entrusted projects, show that the proposed algorithm is faster and more reliable than the compared methods in target detection for serving 3-D registration.

Mesh representation of vasculature is fundamental to many medical applications. The benefit is a clean and tidy appearance in terms of visualization, as well as the possibility of applying computer-assisted intervention and preoperative planning for patients. A vasculature mesh is often reconstructed by iso-surfacing its segmented volume data. Clinicians are usually interested in both the vasculature and its centerline. In this paper, we introduce a mesh centerline extraction approach in the case that volume data are unavailable. The extraction method is inspired by an observation that the vasculature is generally composed of piecewise cylindrical shapes. This observation leads to a conceptually simple but effective strategy to tackle the challenging problem of vasculature centerline extraction, which gracefully combines a branch segmentation scheme and a series of advanced techniques in discrete geometry processing. Our method competes favorably with three state-of-the-art methods in the completeness and accuracy of the extracted centerlines from real human vessels, including the pathological vasculature. Our method also usually leads to maximal and mean extraction errors of less than 1% and 0.5%, respectively.

An accurate temperature profile of the multi-stage fractured horizontal well is the foundation of production profile interpretation using distributed temperature sensing. In this paper, an oil-water two-phase flow multi-stage fractured horizontal well temperature prediction model considering stress sensitivity effect and the Joule–Thomson effect is constructed. Based on the simulation calculation, the wellbore temperature variation under different formation parameters, water cuts, and fracture parameters is discussed. The wellbore temperature distribution in multistage fractured horizontal wells is affected by many factors. According to the principle of orthogonal experimental design, the difference between wellbore temperature and initial formation temperature is selected as the analysis condition. Sixteen groups of orthogonal experimental calculations are designed and conducted. By analyzing the experimental results, it is found that the fracture half-length, water production, and formation permeability are the main controlling factors of the wellbore temperature profile. Finally, the production profile of the well is determined by calculating the temperature profile of a tight oil well and fitting it to the measured data of distributed temperature sensing.