Ensuring the safety of autonomous bulldozers in open-pit mines requires accurate detection of hazardous obstacles such as sumps, stones, and hollows. However, detecting these hazards is challenging due to low contrast with the background, blurry boundaries, and variations in texture, shape, size, and distribution. To address these challenges, this paper proposes Segment Anything Model based mine hazardous obstacle detection model (Mine-SAM). Mine-SAM enhances segmentation performance for hazardous objects through the Weighted Mixture Adapters structure (WMix adapter), Dual Attention Mechanism (DAT), and Wavelet convolutions-based Receptive Field Blocks (WT_RFB). On a self-constructed dataset spanning multiple scenes, Mine-SAM achieved a mean Intersection over Union (mIoU) of 0.9158 and maintained an inference speed exceeding 12 FPS (Frames Per Second) for video. Analysis of detection results across adverse operational scenarios and construction stages demonstrates the model's stability and reliable detection performance, highlighting its practical value in ensuring the safety of autonomous bulldozers during operation.

Graph Neural Networks (GNNs) have been widely applied to software defect prediction with demonstrated promising performance. However, most existing approaches rely on simple graph representations of source code, making them inadequate for capturing high-order interaction patterns among multiple code entities and often overlook complex non-linear semantic and structural dependencies inherent in software systems. Furthermore, as software systems evolve dynamically across versions, inter-entity relationships undergo continuous changes, yet such cross-version evolutionary patterns are rarely considered in existing methods. To address these limitations, we propose a novel software defect prediction approach based on Temporal Hypergraph Neural Networks(THGNN2defect). We first parsed the source code into abstract syntax trees (ASTs) and extracted semantic features using convolution neural networks. Afterward, a semantic hypergraph based on semantic similarity among code entities and a structural hypergraph based on topological relationships in the class dependency graph are constructed, respectively. The extracted semantic and structural features are then combined to initialize node properties in the hypergraphs. Building upon this, semantic and structural hypergraphs from different versions are aligned and connected along a temporal axis to construct a temporal hypergraph that captures the dynamic evolution of code over time. Finally, we introduced a temporal hypergraph convolution and aggregation mechanism to integrate multi-version information, and derive the final semantic and structural representations of the code for defect prediction. We evaluated the proposed method on seven open-source projects by comparing with seven baseline approaches. The experimental results validate its effectiveness, showing that THGNN2defect achieves an average improvement of $\mathbf{2 \%} \boldsymbol{\sim} \mathbf{2 7. 1 \%}$ in F1-score and $\mathbf{1. 7 \% ~ 1 4. 2 \%}$ in AUC compared to the baselines.

Accurate identification of cloud phase in the Arctic is critical for evaluating surface energy budgets and reducing uncertainties in climate models, as clouds exert a complex, warming influence highly sensitive to phase partitioning amidst amplified warming. The submillimeter and infrared radiation exhibit distinct sensitivities toward hydrometeors in different cloud phases. In this study, the performance of synergistic observations of submillimeter and infrared spectrum for cloud identification and phase classification in the Arctic is explored, through sensitivity analysis and classification accuracy analysis. The spectral variances between the submillimeter and infrared bands under the scenarios of clear sky, ice cloud, liquid water cloud, and mixed-phase cloud are analyzed by quantifying the disparities in the observed spectra. The sensitivity analysis reveals that the synergistic metrics constructed by synergistic channels can distinguish clouds from clear skies or identify cloud thermodynamic phases. To quantitatively estimate the classification performance of the combined spectrum, a synergistic classification algorithm is constructed based on the Random Forest framework, and then trained and tested by the simulated synergistic observation datasets in the Arctic. Results from assessment metrics revealed that the overall accuracy of the classification model reaches 91.35%. Especially for clear skies and ice clouds, the classification accuracy is 99.43% and 92.23%, respectively. While for mixed-phase clouds, the overall classification accuracy reaches 87.34%. Specifically, ice-over-water clouds demonstrate 89.40% classification accuracy, while water-over-ice clouds achieve lower accuracy of 85.52%, reflecting fundamental differences in their thermodynamic stability and radiometric signatures. Results provide a robust statistical foundation for advanced cloud detection and phase classification algorithms, demonstrating notable improvements in classification accuracy by synergistic channels. With the upcoming spaceborne submillimeter and infrared passive sensors, the results demonstrate a pressing need and the potential of combining observations to better understand cloud phase and evolution in the future.

Atmospheric temperature and humidity profiles are the basic parameters used to describe the vertical distribution of atmospheric states. Continuous observations of accurate temperature and humidity profiles are essential for exploring boundary layer thermal and dynamic characteristics. To this end, an intelligent retrieval algorithm (IReA) based on a convolutional neural network (CNN) is proposed to retrieve atmospheric temperature and humidity profiles by combining observations from ground-based infrared hyperspectral radiometers and microwave radiometers (MWRs). The results show that the inclusion of microwave observations can effectively improve the retrieval accuracy of temperature and humidity profiles relative to the results from atmospheric emitted radiance interferometer (AERI) under clear-sky conditions, where the root mean square error (RMSE) of the temperature profile is 0.79 K and the RMSE of the humidity profile is 0.95 g/kg. The accuracies of different retrieval methods are also evaluated. In general, the RMSE derived from IReA is improved by at least 9% compared to the results from the physical retrieval method and BP neural network method. Given that clouds are semitransparent in the microwave region, the retrieval accuracy of the temperature and humidity profile of IReA are also improved under cloudy conditions when microwave observations are introduced.

The submillimeter and microwave radiometers exhibit distinct sensitivities toward hydrometeor in different cloud phases. This work aimed to investigate the application of the synergistic observations of spaceborne submillimeter and microwave radiometers in cloud detection and thermodynamic phase classification. The spectral variances between the submillimeter and microwave bands under the scenarios of clear sky, ice-phase cloud, liquid-phase cloud, and mixed-phase cloud were highlighted by quantifying the disparities in the observed spectra. Cloud detection and phase classification models based on random forests were trained and tested by the simulated synergistic observation datasets. Results from assessment metrics revealed that the overall accuracy of the classification model incorporating all features was 88.75%. Feature importance analysis provided the key metrics for cloud phase classifications, classification models trained with these simplified combinations of channels, and metrics achieved an accuracy of 86.25%. The results demonstrated the feasibility of combining submillimeter and microwave radiometer observations for detection and phase classification.

Hyperspectral image (HSI) denoising is critical for the effective analysis and interpretation of hyperspectral data. However, simultaneously modeling global and local features is rarely explored to enhance HSI denoising. In this letter, we propose a hybrid convolution and attention network (HCANet), which leverages both the strengths of convolution neural networks (CNNs) and Transformers. To enhance the modeling of both global and local features, we have devised a convolution and attention fusion module aimed at capturing long-range dependencies and neighborhood spectral correlations. Furthermore, to improve multi-scale information aggregation, we design a multi-scale feed-forward network to enhance denoising performance by extracting features at different scales. Experimental results on mainstream HSI datasets demonstrate the rationality and effectiveness of the proposed HCANet. The proposed model is effective in removing various types of complex noise. Our codes are available at https://github.com/summitgao/HCANet.

Improving the accuracy of measuring ice cloud properties is crucial for the study of atmospheric circulation and climate models, and for understanding the radiative forcing effects of ice clouds. In this study, a novel approach to retrieve ice cloud microphysics involving synergistically analyzing the spectra of submillimeter (sub-mm) and infrared (IR) is proposed, combining the complementary information regarding ice cloud properties from each spectrum. The sensitivity of the synergistic channel pairs to ice water paths (IWPs) and mean mass diameters is thoroughly investigated by the synthetic lookup tables and sensitivity parameter analysis. A synergistic retrieval algorithm based on Quantile Regression Neural Networks is constructed toward a better evaluation of the retrieval biases and uncertainties quantitatively. The simulated retrieval results reveal that the synergistic retrievals outperform the results from individual spectra across the full range of IWP from 1 to 1000 g/m2 and mean mass diameter from 1 to$500~\mu $m. Specifically, the mean root-mean-square-error of the synergistic retrievals for IWP is 68% (95%) lower than that of the sub-mm-only (IR-only) retrievals, and a 10% (24%) lower root-mean-square-error for mean mass diameter, respectively. In addition, the synergy can improve the correlation for IWP by 3.7% (5.6%) and yields a 12.5% (17.6%) higher correlation for mean mass diameter compared to the sub-mm-only (IR-only) retrievals. With the upcoming spaceborne sub-mm and IR passive sensors, the results demonstrate a pressing need and the potential of combining observations to better understand ice cloud properties in the future.

Removing noise from hyperspectral images (HSIs) has been widely regarded as one of the most meaningful preprocessing tasks in remote sensing image interpretation. In this paper, we aim to extend the Transformer backbone to HSI denoising, and propose a Multi-scale Transformer Denoising Network (MTDNet). Specifically, we design a multi-head global attention module to alleviate the computational burden caused by self-attention. Furthermore, we propose a multi-scale feed-forward network in which three branches of multi-scale features are extracted through dilated convolution. It enriches the non-linear feature transformation in the Transformer block. Both the objective and subjective experiments on the ICVL dataset demonstrate the superiority of the proposed MTDNet over four closely related methods.

Building agents based on tree-search planning capabilities with learned models has achieved remarkable success in classic decision-making problems, such as Go and Atari.However, it has been deemed challenging or even infeasible to extend Monte Carlo Tree Search (MCTS) based algorithms to diverse real-world applications, especially when these environments involve complex action spaces and significant simulation costs, or inherent stochasticity.In this work, we introduce LightZero, the first unified benchmark for deploying MCTS/MuZero in general sequential decision scenarios. Specificially, we summarize the most critical challenges in designing a general MCTS-style decision-making solver, then decompose the tightly-coupled algorithm and system design of tree-search RL methods into distinct sub-modules.By incorporating more appropriate exploration and optimization strategies, we can significantly enhance these sub-modules and construct powerful LightZero agents to tackle tasks across a wide range of domains, such as board games, Atari, MuJoCo, MiniGrid and GoBigger.Detailed benchmark results reveal the significant potential of such methods in building scalable and efficient decision intelligence.The code is available as part of OpenDILab at https://github.com/opendilab/LightZero.

Fetal heart rate (fHR) is critical for assessing fetal health and diagnosing disorders, such as fetal distress, congenital heart disease, and intrauterine growth retardation. With the rapid development of the Internet of Medical Things (IoMT), fetal R-peak detection plays an important role in diagnosing heart defects during pregnancy. However, due to the nonlinear mixing of multiple sources in the noninvasive signals and the low signal-to-noise ratio (SNR), it is difficult to obtain accurate R-peak detection result. This article presents a dual self-calibrating system based on a spectral attention kernel independent component analysis (SA-KICA) module and a self-calibrating fetal R-peak detection (SC-FRD) module. SA-KICA is an ICA-based calibration module constructed by the spectral attention mechanism, which was sought from short-time Fourier transform (STFT) and was shipped back to original signal with convolution to achieve perfect maternal electrocardiogram (MECG) separation in high-dimensional linear separable space. Then, a periodic and morphological-based channel selector is designed to select the optimal MECG. After MECG removal, to further improve the performance of fetal R-peak detection, the SC-FRD module is introduced to utilize the interior peak information and self-calibrating strategy, which includes variance-based fetal R-peak seed selection, time-varying coarse prediction, and adaptive probability mask calibration. The proposed framework is a primary attempt to concurrently introduce the nonlinear feature, spectral information, and self-calibrating strategy in the field of fetal ECG processing. The framework achieved excellent performance in fetal R-peak detection accuracy on a simulated data set and two public data sets with varying divergence and richness of resources. The experimental results show that our framework is superior to existing methods and can be used as a potential fetal monitoring method in the application of IoMT. The code is released inhttps://github.com/bfyjr/NI-FECG-Extraction.

Hybrid low-density parity-check (LDPC) decoding combines conventional Belief-Propagation (BP) algorithm with stochastic decoding to achieve high performance and low complexity simultaneously. However, lossy and inefficient stochastic-to-binary (S2B) conversion brings extra performance degradation and decoding latency. In this paper, a bit-serial stochastic updating based hybrid decoding (BSSU-HD) is proposed, which employs fully correlated stochastic (FCS) check nodes (CNs) and probability tracers assisted variable nodes (VNs) to accomplish accurate and efficient S2B conversion. Two strategies, including random source selection and tracing speed switching, are proposed to further improve performance and convergence. A BSSU LDPC decoder for IEEE 802.3an is designed in a 65-nm CMOS process, which occupies 4.6 mm2 silicon area and achieves a throughput of 200.8 Gb/s at$E_{b}/N_{0} = 4.4$dB with 500 MHz clock frequency from a 1.2 V supply voltage. The power and energy efficiency are 2.933 W and 14.61 pJ/bit, respectively. To the best of our known, it achieves the best decoding performance, the highest throughput and hardware efficiency among state-of-the-art IEEE 802.3an LDPC decoders. We also verify that the BSSU-HD can achieve better performance for multi-rate 5th generation (5G) New Ratio (NR) LDPC codes than conventional algorithm, which greatly extends the application of the stochastic decoding.

A retrieval methodology based on the Bayesian neural network (BNN) is presented that inverts the ice water path (IWP), mean mass-weighted diameter (Dme), and cloud height of ice clouds from sub-millimeter radiometer observations. The training dataset was created using collecting cloud profiles from the DARDAR (raDAR/liDAR) database and running simulations by the Atmospheric Radiative Transfer Simulator (ARTS) model. Since the effective radius (re) is the size descriptor of ice particles in the DARDAR database, a look-up table of ice water content (IWC), Dme, and rewas constructed to convert reprofiles into Dmeprofiles. In addition, random noises corresponding to the measurement uncertainties of the Compact Scanning Submillimeter-wave Imaging Radiometer (CoSSIR) during the TC4 experiment were added to the simulated brightness temperatures before training the BNN. The proposed retrieval method was first applied to the simulated testing database, and then to the observations of CoSSIR. Moreover, the retrieved IWP and Dmewere compared to the retrievals of the Bayesian Monte Carlo Integration (BMCI) method. The retrieved cloud height was assessed by cloud height extracted from the reflectivity data of the Cloud Radar System (CPS) flow on the same aircraft with CoSSIR. The comparison showed that the correlation coefficients of the retrieved IWP and Dmefrom the two methods are above 0.8, and the retrieved cloud height also showed good agreement with that extracted from the CPS.

Accurately determining the height of the planetary boundary layer (PBL) is important since it can affect the climate, weather, and air quality. Ground-based infrared hyperspectral remote sensing is an effective way to obtain this parameter. Compared with radiosonde measurements, its temporal resolution is much higher. In this study, a method to retrieve the PBL height (PBLH) from the ground-based infrared hyperspectral radiance data is proposed based on machine learning. In this method, the channels that are sensitive to temperature and humidity profiles are selected as the feature vectors, and the PBLHs derived from radiosonde are taken as the true values. The support vector machine (SVM) is applied to train and test the data set, and the parameters are optimized in the process. The data set collected at the Atmospheric Radiation Measurement (ARM) program Southern Great Plains (SGP) from 2012 to 2015 is analyzed. The instruments used in this letter include Atmospheric Emitted Radiance Interferometer (AERI), Vaisala CL31 ceilometer, and radiosonde. It shows that the root mean square error (RMSE) between the PBLHs calculated by the proposed method using AERI data and those from radiosonde data can be within 370 m, and the square correlation coefficient (SCC) is greater than 0.7. Compared with the PBLHs derived from the ceilometer, it can be found that the new method is more stable and less affected by clouds.

The ultra-low hardware consumption feature of stochastic decoding has made it a potential candidate for the implementation of low-density parity-check(LDPC) decoders. However, the existing stochastic LDPC decoders still suffer from performance degradation and relatively high decoding cycles caused by the correlation among stochastic bit streams. In this paper, we propose Hybrid Stochastic(HS) decoding, which achieves high performance, high throughput, and high hardware efficiency by jointly using our proposed novel stochastic check node(CN) and Two’s Complement(TCS) variable node(VN) to realize Min-Sum Algorithm(MSA) and its enhancements. Fully correlated stochastic bit streams are used to entirely eliminate the indeterminacy caused by the correlation, which results in high performance and fast convergence and inherits the low complexity of stochastic decoders at the same time. We demonstrate the HS decoding by designing a (2048,1723) decoder in a 65 nm process, which achieves the highest Bit-Error-Ratio(BER) performance, highest throughput, and top hardware efficiency among existing stochastic LDPC decoders. We also demonstrate that HS decoding can achieve excellent decoding performance for different code rates and lengths 5G New Radio(NR) LDPC codes. Thus, HS decoding can be adopted in wide applications.

MOTIVATION: Combination therapy is widely used in cancer treatment to overcome drug resistance. High-throughput drug screening is the standard approach to study the drug combination effects, yet it becomes impractical when the number of drugs under consideration is large. Therefore, accurate and fast computational tools for predicting drug synergistic effects are needed to guide experimental design for developing candidate drug pairs. RESULTS: Here, we present TAIJI, a high-performance software for fast and accurate prediction of drug synergism. It is based on the winning algorithm in the AstraZeneca-Sanger Drug Combination Prediction DREAM Challenge, which is a unique platform to unbiasedly evaluate the performance of current state-of-the-art methods, and includes 160 team-based submission methods. When tested across a broad spectrum of 85 different cancer cell lines and 1089 drug combinations, TAIJI achieved a high prediction correlation (0.53), approaching the accuracy level of experimental replicates (0.56). The runtime is at the scale of minutes to achieve this state-of-the-field performance. AVAILABILITY AND IMPLEMENTATION: TAIJI is freely available on GitHub (https://github.com/GuanLab/TAIJI). It is functional with built-in Perl and Python. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.