Particle Image Velocimetry (PIV) is widely used for velocity field measurement in multiphase flows because of its non-intrusive operation and high spatial resolution. In gasliquid two-phase flows, however, dense bubbles and irregular bubble morphologies make accurate annotation of instantaneous velocity fields in PIV images difficult, which limits the availability of reliable training datasets for deep learningbased velocity estimation. This study introduces a PIV dataset generation framework that integrates semantic segmentation with Computational Fluid Dynamics (CFD). A semantic segmentation model is employed to extract bubble structures and reconstruct realistic particle distributions. CFD simulations are subsequently used to generate velocity field labels corresponding to bubble locations, enabling physically consistent coupling between flow dynamics and image morphology. Experimental evaluation shows that the generated dataset improves scene fidelity and labeling consistency compared with conventional synthetic PIV datasets. When used to train deep learning-based PIV models, the dataset reduces velocity pseudovector occurrence and noise interference, resulting in measurable improvements in velocity estimation accuracy and robustness under gas-liquid two-phase flow conditions.

In gas-solid two-phase flow, the accuracy of solid concentration measurement is significantly affected by the unknown and varying moisture content in the solid phase, particularly when the used method is sensitive to the moisture content. To address this issue, a new method that uses a novel microwave transmission sensor working at two frequencies and a novel decoupling method is proposed to achieve moisture-independent solid concentration measurement. A dual-frequency resonant printed dipole antenna is designed using the finite element simulation method, and the microwave transmission sensor, which can work at two frequencies, is formed by two dipole antennas. A simple and effective decoupling method is proposed to extract moisture content from dual-frequency phase shift information, enabling solid concentration measurement. The flow structures of roping flow with varying solid concentrations and moisture contents are used to verify the proposed method. Experimental results demonstrate that the dual-frequency phase shifts of the microwave transmission sensor are both sensitive to the solid concentration and moisture content. Therefore, only relying on the phase shift information from a single frequency cannot accurately measure solid concentration under conditions of varying moisture content. The decoupling method proposed in this study, based on dual-frequency phase shift information, can accurately determine moisture content. Consequently, it enables solid concentration measurement in gas-solid flows with relative errors within +/- 5% under conditions of varying moisture content.

The nonuniform solid distribution and unknown solid moisture content in gas-solid two-phase flow present challenges to the measurement of solid concentration. To address this issue, this article proposes a novel method for measuring solid concentration by developing a Yagi antenna-based microwave sensor. The dimensions of the Yagi antenna are optimized through finite element simulation to ensure good radiation performance at the working frequency. The optimized Yagi antennas are employed as transmitting unit and receiving unit, forming a novel microwave transmission sensor. Experiments are carried out to assess the sensor performance and its effectiveness in measuring solid concentrations. Experimental results indicate that the measured phase shift exhibits a high sensitivity to changes in solid concentration and is not sensitive to the nonuniform solid distribution. However, the relation between the measured phase shift and solid concentration is influenced by the moisture content of the solid. In order to eliminate the impact of moisture content on solid concentration measurement, a new parameter named normalized phase shift is defined. The relation between the solid concentration and the normalized phase shift is almost not affected by the moisture content. Consequently, the Yagi antenna-based microwave sensor, combined with the definition of normalized phase shift, allows for accurate measurement of solid concentration under varying moisture conditions.

Objective. Accurate polyp segmentation from colo-noscopy images plays a crucial role in the early diagnosis and treatment of colorectal cancer. However, existing polyp segmentation methods are inevitably affected by various image noises, such as reflections, motion blur, and feces, which significantly affect the performance and generalization of the model. In addition, coupled with ambiguous boundaries between polyps and surrounding tissue, i.e. small inter-class differences, accurate polyp segmentation remains a challenging problem. Approach. To address these issues, we propose a novel two-stage polyp segmentation method that leverages a preprocessing sub-network (Pre-Net) and a dynamic uncertainty mining network (DUMNet) to improve the accuracy of polyp segmentation. Pre-Net identifies and filters out interference regions before feeding the colonoscopy images to the polyp segmentation network DUMNet. Considering the confusing polyp boundaries, DUMNet employs the uncertainty mining module (UMM) to dynamically focus on foreground, background, and uncertain regions based on different pixel confidences. UMM helps to mine and enhance more detailed context, leading to coarse-to-fine polyp segmentation and precise localization of polyp regions. Main results. We conduct experiments on five popular polyp segmentation benchmarks: ETIS, CVC-ClinicDB, CVC-ColonDB, EndoScene, and Kvasir. Our method achieves state-of-the-art performance. Furthermore, the proposed Pre-Net has strong portability and can improve the accuracy of existing polyp segmentation models. Significance. The proposed method improves polyp segmentation performance by eliminating interference and mining uncertain regions. This aids doctors in making precise and reduces the risk of colorectal cancer. Our code will be released at https://github.com/zyh5119232/DUMNet.

An array capacitive sensor adopting two excitation modes is designed for measuring local particle concentration in gas–solid two-phase flow. Considering the difficulty in obtaining instantaneous flow parameters under dynamic experiments, a 3D dynamic simulation model coupling gas–solid two-phase fluid and electrostatic fields is proposed to quantitatively evaluate the sensor performance. In the coupling model, Computational Fluid Dynamics and Discrete Element Method are applied to establish two-phase fluid field for particle micro-distribution. The method of defining particle shape and position is used to set permittivity distribution in electrostatic field for coupling gas–solid fluid field and electrostatic field of array capacitive sensor. Using the method of reconstructed image and data fitting, the performance of two excitation modes are evaluated under a typical dynamic flow with structure of stratified flow. Experimental result validates the simulation conclusions and proves the effectiveness of the coupling simulation model for evaluating capacitive sensors in gas–solid flow systems.

In a gas-solid two-phase flow system, simultaneous measurement of solid concentration and solid moisture is essential for some applications. However, during the measurement process, solid concentration and solid moisture are coupled in the measurement signals when the measurement signal is sensitive to both solid concentration and moisture. This makes it challenging to measure them simultaneously in the same measurement field. To solve this issue, a novel measurement method is proposed in this study, which uses a dual-modality electrical sensor to achieve simultaneous measurement of solid concentration and solid moisture in gas-solid two-phase flow. The dual-modality electrical sensor features a simple structure and captures measurement signals of capacitance and angle of dielectric loss within the same measurement section. With the above signals which are both sensitive to solid concentration and moisture, an iterative correction calculation method is adopted to achieve the simultaneous measurement of solid concentration and solid moisture. During the calculation process, the measured parameters are decoupled and continuously corrected until the related error values meet the threshold requirement. Numerical simulation experiments under three typical distributions and physical experiments under roping flow are carried out to verify the proposed measurement method. The results prove the effectiveness of the proposed measurement method in the simultaneous measurement of solid concentration and solid moisture for gas-solid two-phase flow.

Capacitive solid concentration measurement has been widely applied in gas-solid two-phase flow. In the existing method, the solid concentration measured from capacitance is based on a constant solid permittivity. However, due to factors such as the raw material ratio and moisture content, solid permittivity may vary during the measurement process, and in some cases even significantly. A novel concentration measurement method using a combination of capacitive sensors is proposed, in which a calibration sensor operating at a certain solid concentration provides solid permittivity information online, and then the output of the calibration sensor is used to correct the solid concentration in measurement sensor. An online calibration algorithm is presented based on the relationship between the outputs of two capacitive sensors, in which the solid permittivity information is given directly by the output of calibration sensor without the calculation of the solid permittivity. Moreover, it can obtain the relationship between the solid concentration and the output of the measurement sensor for the arbitrary solid permittivity over the working range. Static and dynamic experiments in the laboratory were carried out to verify the effectiveness of the proposed method. The average relative measurement error of average solid concentration was reduced from 34.07% to 6.68% when the solid permittivity information ranged from 2.09 to 5.46 in the dynamic experiments. The proposed method has been applied in the industrial tests of pulverized coal injection system of blast furnace, and the measured one-hour accumulated mass of pulverized coal can better agree with the true value.

Global contexts are critical to locating salient objects for salient object detection (SOD). However, the convolution operation in CNNs has a local receptive field, which cannot capture long-distance global information. Recent studies have shown that modernized CNN models with large kernel convolution, such as ConvNeXt, can effectively extend the receptive fields. Based on it, this paper explores the potential of large kernel CNN for SOD task. Inspired by the common information between RGB and depth images in salient objects, we propose a ConvNeXt-based Siamese network with shared weight parameters. This structural design can effectively reduce the number of parameters without sacrificing performance. Furthermore, a depth information preprocessing module is proposed to minimize the impact of low-quality depth images on predicted saliency maps. For cross-modal feature interaction, a dynamic fusion module is designed to enhance cross-modal complementarity dynamically. Extensive experiments and evaluation results on six benchmark datasets demonstrate the outstanding performance of the proposed method against 14 state-of-the-art RGB-D methods. Our code will be released at https://github.com/zyh5119232/CSNet .

Current vertebral bone segmentation algorithms exhibit limitations, including insensitivity to anatomical variations, proneness to overfitting with limited training data, and susceptibility to errors caused by noise and interference. In this paper, a fully automated deep learning method is proposed for vertebral body localization and segmentation. The method utilizes Spatial Configuration-Net and U-Net. Gaussian blur and image intensity clamping are initially applied to the CT images to reduce noise resulting from screws and implants. Subsequently, U-Net estimates the position of the spinal column, followed by vertebral body localization and identification achieved through the application of Spatial Configuration-Net (SCN) and heatmap regression. SCN incorporates prior knowledge of the spatial configuration of labeled vertebrae, thereby enhancing sensitivity to anatomical variations. To mitigate the overfitting issue, measures such as increasing training data and introducing regularization techniques are implemented. The proposed deep learning model has valuable implications for applications such as pedicle screw placement surgery that necessitate object localization and segmentation. Comparative validation tests conducted against other models highlight significant advantages and practical value offered by the proposed model.

Existing RGB-D SOD methods mainly rely on a symmetric two-stream CNN-based network to extract RGB and depth channel features separately. However, there are two problems with the symmetric conventional network structure: first, the ability of CNN in learning global contexts is limited; second, the symmetric two-stream structure ignores the inherent differences between modalities. In this paper, we propose a Transformer-based asymmetric network (TANet) to tackle the issues mentioned above. We employ the powerful feature extraction capability of Transformer (PVTv2) to extract global semantic information from RGB data and design a lightweight CNN backbone (LWDepthNet) to extract spatial structure information from depth data without pre-training. The asymmetric hybrid encoder (AHE) effectively reduces the number of parameters in the model while increasing speed without sacrificing performance. Then, we design a cross-modal feature fusion module (CMFFM), which enhances and fuses RGB and depth features with each other. Finally, we add edge prediction as an auxiliary task and propose an edge enhancement module (EEM) to generate sharper contours. Extensive experiments demonstrate that our method achieves superior performance over 14 state-of-the-art RGB-D methods on six public datasets. Our code will be released at https://github.com/lc012463/TANet.

In gas-solid two-phase flow, moisture in solid brings new challenge to the accurate measurement of solid concentration. A combined measurement method is proposed in this paper, which realizes the measurement of moisture and solid concentration in the gas-solid two-phase flow. The combined measurement is composed of capacitance measurement method and displacement-current phase (DCP) measurement method, which are both sensitive to moisture and solid concentration. The DCP measurement based on the principle of dielectric loss in measured field can share the same sensor with the capacitance measurement, which ensures the same measured section in the combined measurement method. Based on the combined measurement signal, the iterative decoupling method is adopted to realize the accurate measurement of moisture and solid concentration at same time, where the relationship between measurement and solid concentration for the expected moisture can be updated. The proposed method is verified by simulation experiments, where the results show that it can effectively obtain moisture and solid concentration in the gas-solid two-phase flow.

Existing RGB-D salient object detection methods generally rely on the dual-encoder structure for RGB and depth feature extraction. However, we observe that the encoders in such models are often not adequately trained to obtain superior feature representations. We name this problem the “under-training issue”. To this end, we propose a multi-branch decoding network (MBDNet) to suppress this issue. The MBDNet introduces additional decoding branches with supervision to form a multi-branch decoding (MBD) structure, facilitating the training of the encoders and enhancing the feature representation. Specifically, to ensure the effectiveness of the introduced supervision and improve the performance of additional decoding branches, we design an adaptive multi-scale decoding (AMSD) module. We also design a multi-branch feature aggregation (MBFA) module to aggregate the multi-branch features in MBD to further improve the detection accuracy. In addition, we design an enhancement complement fusion (ECF) module to achieve multi-modality feature fusion. Extensive experiments demonstrate that our MBDNet outperforms other state-of-the-art methods and mitigates the “under-training issue”.

Gas-solid flows with nonuniform phase distribution and varying moisture often exist in the industrial process. Its solid concentration measurement is an important and challenging research field, and new effective measurement methods are needed. In this study, a novel method using the microwave resonant cavity sensor is proposed to measure solid concentration of gas-solid two-phase flow with varying moisture. The sensor is developed from a cylindrical resonant cavity with open ends, and the orthogonal experiment is carried out using simulation method to determine the resonant mode and optimize the sensor configuration. Afterward, experiments are conducted to evaluate the performance of the designed sensor and the proposed method. The sensor has a high and nearly constant sensitivity to the solid concentration and its output is almost not affected by the nonuniform phase distribution. The varying moisture brings effect to solid concentration measurement when the proposed sensor is used. To solve this problem, the new parameter named normalized resonant frequency is defined to reduce this influence, and finally, the solid concentration measurement is achieved.

Simultaneous localization and mapping (SLAM) technology is one of the core technologies of the crawler robot with integrated inspection and firefighting. Due to the existence of degraded environment in the cable tunnel, it leads to a large cumulative error of the system. The correct loop closure detection(LCD) can well eliminate the cumulative error, which is of great significance to the intelligent inspection of the robot. Aiming at the problems of low recall rate of LCD and inability to construct globally consistent trajectories and maps for crawler inspection robots due to similar and single features in cable tunnel environment, a LCD method based on ArUco markers is proposed. In the image feature extraction, the image feature extraction method is designed according to the environmental characteristics of cable tunnel, which makes the image features rich and differentiated; On the construction of loop back key frame, a loop back key frame cluster library for loop back detection is constructed according to the unique ID information of ArUco, which solves the problem that the correct loop back candidate frame cannot be detected in similar environment; in the research of loop detection algorithm, ArUco mark is introduced for loop verification, which improves the recall rate of loop detection. The experimental results show that the method can accurately achieve LCD in the cable tunnel environment. It improves the autonomous inspection accuracy of the inspection robot under the cable tunnel.

In gas–solid two-phase flows, the situation that moisture in the solid changes is often encountered. The varying and unknown moisture brings new challenges to the solid concentration measurement if the used measurement method is sensitive to moisture. In this study, a new method that can eliminate the influence of moisture on the measurement of solid concentration in gas–solid flows is proposed. First, the new combined sensors consisting of the microwave transmission sensor and the capacitance sensor are proposed, and the measurement systems are designed. Both the microwave transmission sensor and the capacitance sensor are sensitive to moisture and solid concentration, and they can acquire information about the solid concentration and moisture simultaneously. Then an effective decoupling method is proposed to derive moisture and solid concentration from the information acquired by the microwave transmission sensor and the capacitance sensor. Experiments are carried out to simulate the flow structure of roping flow to verify the proposed method. The results indicate that the influence of moisture on the measurement of solid concentration can be eliminated, and the moisture-independent measurement of solid concentration in gas–solid flows can be realized.

The cross temperature measuring device is applied to monitor the top temperature and gas flow distribution in the blast furnace (BF) during the ironmaking process. However, the temperature of the BF roof is relatively high, which causes damage to the cross temperature measuring device. Therefore, this paper proposes an efficient estimation method. First of all, the data filtering method are used to solve the problems of noise interference. Moreover, as the fact that BF cross temperature measurement is affected by interference factors, it is easy to cause information redundancy. Aiming at this, the maximum information coefficient (MIC) correlation analysis and sequence dislocation method are used to determine the input variables and its action time of the model. Finally, an estimation model based three-layer long and short-term memory network (LSTM3) is established. The experimental results show that the LSTM3 can estimate the centre temperature of cross temperature measuring device.

Multi-electrode excitation is an effective way to improve the performance of 3D electrical tomography (ET) system compared with single electrode excitation. This paper systematically discusses various sensing strategies for multi-electrode excitation with different combinations in radial and axial directions. A typical 3 x 8 3D ERT sensor is adopted to analyze the influence of different excitation modes on performance indexes, such as number of independent measurements, dynamic range of measurements, sensitivity distribution and correlation coefficient of image reconstruction. On the basis of radial rotation angle between the excitation electrodes mapped to the same plane and whether the excitation electrodes are located in the same axial layer, five typical electrode combination modes are designed under dual-electrode excitation and single-electrode measurement protocol. The results show that the variation trend of different indexes influenced by excitation mode is inconsistent. The entropy-based method is applied to comprehensively evaluate the different excitation modes with the selected multiple performance indexes. Among all above modes, the mode in which excitation electrodes are vertically aligned is better universal one. The experimental results show that the position characteristics of the preferable mode can obtain better image reconstruction quality in different distribution models. Furthermore, the better excitation mode will provide guidance for design of other multi-electrode excitation in 3D ET systems with different structures.

RGB-D Salient Object Detection (SOD) methods utilize the information provided by the depth map to assist detection. To make effective use of depth information, existing methods face two challenges: (1) How to obtain valuable cross-modal complementary information. (2) How to reduce the negative effects of unreliable depth maps for RGB-D SOD. To prevent unreliable depth maps from interfering with the detection results, we design a feature enhancement module (FEM). The module first uses a feature guidance unit (FGU) to guide the depth feature to restore the saliency information, and then uses a feature selection unit (FSU) to obtain the higher quality depth feature. Finally, the representation of depth features is deeply enhanced through progressive steps of guidance and selection. To obtain valuable cross-modal complementary information, we design a multi-scale cross-modal attention module (MCA). We first use multiple parallel dilated convolutional layers to capture global and local information, and then we design a cross-modal attention unit (CAU) to generate cross-modal attention maps. These maps can eliminate redundant information and suppress background interference. Quantitative and qualitative experiments on five datasets demonstrate the superiority of the proposed approaches.

Accurate detection for pulverized coal injection (PCI) blockage and coke size distribution are effective for blast furnace (BF) operation. Nowadays, tuyere cameras have been applied in BF. However, the appearance of captured raceway images changes, which requests the detection method to be adaptive. Therefore, an intelligent detection method for PCI blockage and coke size distribution is proposed. An adaptive pre-processing algorithm is firstly developed to improve image quality. Secondly, the fitting circle is used to locate tuyere region and the improved U-net network is investigated for lance segmentation to construct the background template. Then, target regions can be obtained by background subtraction algorithm. Finally, the PCI blockage can be detected according to the area information, and the size distribution of cokes can be calculated by multi-directional Feret diameter. Massive raceway videos are used to evaluate the method. Experiment results show the method can detect PCI blockage and cokes size distribution.

Recently, most CNNs-based Salient Object Detection (SOD) models have achieved great progress through various feature aggregation strategies. But most of them usually introduce a large number of features into a module for fusion, which results in information dilution. In this paper, we propose a Progressive Fusion Network (PFNet) to solve this problem. The PFNet alleviates the information dilution through a Progressive Fusion Architecture (PFA), which aggregates the features extracted from encoder in a progressive fusion manner. In addition, we use a simple Feature Fusion Module (FFM) that utilize high-level features to enhance the semantic information of low-level features, thereby ensuring the effective fusion of features. Finally, we leverage an Enhanced Loss to guide the optimization process of the network and obtain high-quality saliency maps. The whole network is trained end-to-end without any pre-processing and post-processing. The quantitative and qualitative experimental results on five benchmark datasets demonstrate that the superiority of the proposed approaches.