Deep Network Research Articles

Tool wear prediction in milling CFRP with new fiber orientation based on deep feature transfer learning network

ABSTRACT The machining of carbon fiber reinforced polymers (CFRP) is an essential step in converting the near-shape component to its final geometry in the manufacturing chain of CFRP. However, tool wear grows rapidly because of the highly abrasive property of carbon fibers, resulting in surface damage and poor surface roughness. This paper proposed a deep feature transfer learning network for tool wear prediction in milling undirectional (UD) CFRP with new fiber orientation by considering different features. The features are divided into transferable common features about the increase of tool wear and unique features about the distribution of force signals. The proposed method learns the common features of the tool wear progression process from two different historical cutting conditions. The loss calculation and parameters iteration techniques are utilized, which can increase the applicability for different cutting parameters. Besides, the common feature transfer and unique feature learning strategy remain the common features, and learn the unique features about the specific cutting parameters and fiber orientation in new cutting conditions. The effectiveness of the proposed method is experimentally validated with different transfer prediction tasks. The proposed method’s advantages in comparison with other transfer learning and non-transfer learning methods are demonstrated.

From courtyard to atrium: spatial differentiation in the spontaneous evolution of vernacular architecture and its response to geo-climate

ABSTRACT Vernacular architecture is deeply rooted in specific regions and evolves under urbanization while maintaining a close connection to the natural environment. Using the case study of vernacular courtyards evolving into vernacular atriums, this study examines the spatial distribution characteristics of vernacular atriums in 37 counties (districts) in southern Hebei, China, by mining data through deep learning networks. The results reveal a tendency for vernacular atriums to concentrate in the southwest, while regions less frequented, such as the east and north, hold potential for promotion. Furthermore, the correlation between 15 geo-climatic factors and vernacular atriums is explored. Geodetector analysis indicates that altitude, slope, rainfall, wind speed, sunshine duration, shortwave radiation intensity and PM2.5 provide varying levels of explanatory power for different types of vernacular atriums. The response of different factors reflects the similarities and differences in the needs of various vernacular atriums to improve the living environment and adapt to the climate conditions, such as insulation, sunshade, lighting, rain protection, and ventilation, while adapting the geographical environment through house form and culture. With AI-assisted fieldwork, this study offers insights into the macro-scale climate adaptability of vernacular courtyards, inspiring sustainable development in vernacular architecture.

DSF-Net: semantic segmentation of large-scale point clouds based on integrating deep and shallow networks

PurposeWith the upgrading of three-dimensional (3D) sensing devices, the amount of point cloud data collected has also increased exponentially. However, most of the existing methods also have unbalanced optimizations in memory consumption and semantic segmentation efficiency. This research addresses the need for a more balanced approach in processing large-scale point cloud data efficiently.Design/methodology/approachThis research used a network framework (DSF-Net) based on dual-path deep and shallow networks and designed a point cloud space pyramid pooling module based on hole convolution. The 3D point cloud data are trained separately by integrating the deep branch and shallow branch networks. Besides, a deep and shallow fusion module fuses the deep and shallow feature relationships and outputs several loss functions for convergence training.FindingsIt is found that DSF-Net solves the problem of segmentation efficiency, achieves a balanced effect while ensuring the ability of a large range of point cloud input and reduces the memory consumption.Originality/valueThe deep network can extract high-level semantic information, while the shallow neural network has fewer neural network layers and faster inference speed. Meanwhile, random sampling and point-atrous spatial pyramid pool modules are used, respectively, for deep and shallow networks to capture multi-scale local context information in point cloud.

CoLeafNet: a dual-track deep learning network for classification of nutrient deficiencies in coffee leaves using densenet and efficient multi-scale feature attention network

Abstract Nutrient health is essential for coffee plants as it directly affects their growth, resilience and bean quality. Essential nutrients support key physiological processes in coffee plants and deficiencies can reduce crop yield, compromise bean quality, and lead to economic losses. Given the widespread importance of coffee, especially in developing economies, early detection of nutrient deficiencies in coffee leaves is crucial. However, traditional diagnostic methods like manual inspections and chemical testing are time-intensive, costly and prone to human error. This research work introduces CoLeafNet, a novel dual-track Deep Learning (DL) framework designed specifically for nutrient deficiency classification in coffee leaves. The CoLeafNet architecture integrates DenseNet to extract global, hierarchical features with an Efficient Multi-Scale Feature Attention Network (EMFANet) for detailed local feature capture. EMFANet incorporates an Efficient Channel Attention (ECA) mechanism to enhance cross-channel feature relevance and Ghost modules to minimize computational load while preserving critical information. This combination allows CoLeafNet to dynamically emphasize essential features at various scales, enabling precise detection of nutrient-specific visual patterns. To the best of our knowledge, this is the first dual-track network developed for nutrient deficiency classification in coffee leaves, offering a novel and scalable approach for precision agriculture. It was tested on the CoLeaf dataset, which contains 1,006 labeled images depicting deficiencies in key nutrients like Boron, Iron, Potassium, Calcium, Magnesium, etc. CoLeafNet achieved a classification accuracy of 95.64%, outperforming existing DL models.

Rethinking femoral neck anteversion assessment: a novel automated 3D CT method compared to traditional manual techniques

PurposeTo evaluate the accuracy and reliability of a novel automated 3D CT-based method for measuring femoral neck anteversion (FNA) compared to three traditional manual methods.MethodsA total of 126 femurs from 63 full-length CT scans (35 men and 28 women; average age: 52.0 ± 14.7 years) were analyzed. The automated method used a deep learning network for femur segmentation, landmark identification, and anteversion calculation, with results generated based on two axes: Auto_GT (using the greater trochanter-to-intercondylar notch center axis) and Auto_P (using the piriformis fossa-to-intercondylar notch center axis). These results were validated through manual landmark annotation. The same dataset was assessed using three conventional manual methods: Murphy, Reikeras, and Lee methods. Intra- and inter-observer reliability were assessed using intraclass correlation coefficients (ICCs), and pairwise comparisons analyzed correlations and differences between methods.ResultsThe automated methods produced consistent FNA measurements (Auto_GT: 17.59 ± 9.16° vs. Auto_P: 17.37 ± 9.17° on the right; 15.08 ± 9.88° vs. 14.84 ± 9.90° on the left). Intra-observer ICCs ranged from 0.864 to 0.961, and inter-observer ICCs between Auto_GT and the manual methods were high, except for the Lee method. No significant differences were observed between the two automated methods or between the automated and manual verification methods. Moreover, strong correlations (R > 0.9, p < 0.001) were found between Auto_GT and the manual methods.ConclusionThe novel automated 3D CT-based method demonstrates strong reproducibility and reliability for measuring femoral neck anteversion, with performance comparable to traditional manual techniques. These results indicate its potential utility for preoperative planning, postoperative evaluation, and computer-assisted orthopedic procedures.Clinical trial numberNot applicable.

Enhanced EEG decoding with weight freezing in parallel deep neuro-fuzzy networks (PDNFN) for multi-neurological disorder diagnosis

Enhanced EEG decoding with weight freezing in parallel deep neuro-fuzzy networks (PDNFN) for multi-neurological disorder diagnosis

Smart Home Automation with Smart Metering Using Zigbee Technology and Deep Belief Network

Abstract— This project focuses on developing a smart home automation system that includes smart energy metering using Zigbee technology and Deep Belief Network (DBN). Zigbee provides a low-power, wireless communication method to connect various smart devices in the home. It enables real-time monitoring and control of appliances, lights, and meters through a central controller or smartphone. A smart energy meter is integrated to record electricity usage and help reduce power consumption by providing timely feedback to users. To improve automation and decision-making, a Deep Belief Network is used. This machine learning model learns user behavior patterns and predicts energy usage, allowing the system to automatically manage devices for better energy efficiency and comfort. For example, it can turn off lights when a room is unoccupied or adjust appliance use during peak hours. Overall, this system offers an energy-saving, user-friendly, and intelligent solution for modern smart homes by combining wireless communication and artificial intelligence. Keywords— Smart Home Automation, Smart Metering, Wireless Communication, Home Energy Management

Drainage Network Generation for Urban Pluvial Flooding (UPF) Using Generative Adversarial Networks (GANs) and GIS Data

Mapping urban pluvial flooding (UPF) in data-scarce regions poses significant challenges, particularly when drainage systems are inadequate or outdated. These limitations hinder effective flood mitigation and risk assessment. This study proposes an innovative approach to address these challenges by integrating deep learning (DL) models with traditional methods. First, deep convolutional generative adversarial networks (DCGANs) were employed to enhance drainage network data generation. Second, deep recurrent neural networks (DRNNs) and multi-criteria decision analysis (MCDA) methods were implemented to assess UPF. The study compared the performance of these approaches, highlighting the potential of DL models in providing more accurate and robust flood mapping outcomes. The methodology was applied to Lahore, Pakistan—a rapidly urbanizing and data-scarce region frequently impacted by UPF during monsoons. High-resolution ALOS PALSAR DEM data were utilized to extract natural drainage networks, while synthetic datasets generated by GANs addressed the lack of historical flood data. Results demonstrated the superiority of DL-based approaches over traditional MCDA methods, showcasing their potential for broader applicability in similar regions worldwide. This research emphasizes the role of DL models in advancing urban flood mapping, providing valuable insights for urban planners and policymakers to mitigate flooding risks and improve resilience in vulnerable regions.

Scale Generalisation Properties of Extended Scale-Covariant and Scale-Invariant Gaussian Derivative Networks on Image Datasets with Spatial Scaling Variations

Due to the variabilities in image structures caused by perspective scaling transformations, it is essential for deep networks to have an ability to generalise to scales not seen during training. This paper presents an in-depth analysis of the scale generalisation properties of the scale-covariant and scale-invariant Gaussian derivative networks, complemented with both conceptual and algorithmic extensions. For this purpose, Gaussian derivative networks (GaussDerNets) are evaluated on new rescaled versions of the Fashion-MNIST and the CIFAR-10 datasets, with spatial scaling variations over a factor of 4 in the testing data, that are not present in the training data. Additionally, evaluations on the previously existing STIR datasets show that the GaussDerNets achieve better scale generalisation than previously reported for these datasets for other types of deep networks. We first experimentally demonstrate that the GaussDerNets have quite good scale generalisation properties on the new datasets and that average pooling of feature responses over scales may sometimes also lead to better results than the previously used approach of max pooling over scales. Then, we demonstrate that using a spatial max pooling mechanism after the final layer enables localisation of non-centred objects in the image domain, with maintained scale generalisation properties. We also show that regularisation during training, by applying dropout across the scale channels, referred to as scale-channel dropout, improves both the performance and the scale generalisation. In additional ablation studies, we show that, for the rescaled CIFAR-10 dataset, basing the layers in the GaussDerNets on derivatives up to order three leads to better performance and scale generalisation for coarser scales, whereas networks based on derivatives up to order two achieve better scale generalisation for finer scales. Moreover, we demonstrate that discretisations of GaussDerNets based on the discrete analogue of the Gaussian kernel in combination with central difference operators perform best or among the best, compared to a set of other discrete approximations of the Gaussian derivative kernels. Furthermore, we show that the improvement in performance obtained by learning the scale values of the Gaussian derivatives, as opposed to using the previously proposed choice of a fixed logarithmic distribution of the scale levels, is usually only minor, thus supporting the previously postulated choice of using a logarithmic distribution as a very reasonable prior. Finally, by visualising the activation maps and the learned receptive fields, we demonstrate that the GaussDerNets have very good explainability properties.

Inference-specific learning for improved medical image segmentation.

Deep learning networks map input data to output predictions by fitting network parameters using training data. However, applying a trained network to new, unseen inference data resembles an interpolation process, which may lead to inaccurate predictions if the training and inference data distributions differ significantly. This study aims to generally improve the prediction accuracy of deep learning networks on the inference case by bridging the gap between training and inference data. We propose an inference-specific learning strategy to enhance the network learning process without modifying the network structure. By aligning training data to closely match the specific inference data, we generate an inference-specific training dataset, enhancing the network optimization around the inference data point for more accurate predictions. Taking medical image auto-segmentation as an example, we develop an inference-specific auto-segmentation framework consisting of initial segmentation learning, inference-specific training data deformation, and inference-specific segmentation refinement. The framework is evaluated on public abdominal, head-neck, and pancreas CT datasets comprising 30, 42, and 210 cases, respectively, for medical image segmentation. Experimental results show that our method improves the organ-averaged mean Dice by 6.2% (p-value=0.001), 1.5% (p-value=0.003), and 3.7% (p-value<0.001) on the three datasets, respectively, with a more notable increase for difficult-to-segment organs (such as a 21.7% increase for the gallbladder [p-value=0.004]). By incorporating organ mask-based weak supervision into the training data alignment learning, the inference-specific auto-segmentation accuracy is generally improved compared with the image intensity-based alignment. Besides, a moving-averaged calculation of the inference organ mask during the learning process strengthens both the robustness and accuracy of the final inference segmentation. By leveraging inference data during training, the proposed inference-specific learning strategy consistently improves auto-segmentation accuracy and holds the potential to be broadly applied for enhanced deep learning decision-making.

Fully volumetric body composition analysis for prognostic overall survival stratification in melanoma patients

BackgroundAccurate assessment of expected survival in melanoma patients is crucial for treatment decisions. This study explores deep learning-based body composition analysis to predict overall survival (OS) using baseline Computed Tomography (CT) scans and identify fully volumetric, prognostic body composition features.MethodsA deep learning network segmented baseline abdomen and thorax CTs from a cohort of 495 patients. The Sarcopenia Index (SI), Myosteatosis Fat Index (MFI), and Visceral Fat Index (VFI) were derived and statistically assessed for prognosticating OS. External validation was performed with 428 patients.ResultsSI was significantly associated with OS on both CT regions: abdomen (P ≤ 0.0001, HR: 0.36) and thorax (P ≤ 0.0001, HR: 0.27), with lower SI associated with prolonged survival. MFI was also associated with OS on abdomen (P ≤ 0.0001, HR: 1.16) and thorax CTs (P ≤ 0.0001, HR: 1.08), where higher MFI was linked to worse outcomes. Lastly, VFI was associated with OS on abdomen CTs (P ≤ 0.001, HR: 1.90), with higher VFI linked to poor outcomes. External validation replicated these results.ConclusionsSI, MFI, and VFI showed substantial potential as prognostic factors for OS in malignant melanoma patients. This approach leveraged existing CT scans without additional procedural or financial burdens, highlighting the seamless integration of DL-based body composition analysis into standard oncologic staging routines.Graphical

IHDFN-DTI: Interpretable Hybrid Deep Feature Fusion Network for Drug-Target Interaction Prediction.

Conventional drug discovery is expensive and takes a long period. Drug-target interaction (DTI) prediction through computational methods significantly improves efficiency and reduces costs, holding substantial research value. Despite progress in existing prediction methods, two major challenges remain: first, most methods fail to effectively combine shallow and deep features of protein sequences, overlooking the synergistic effect of both; second, existing feature fusion techniques are relatively simple and struggle to fully capture the complexity and richness of fused features. We suggest an interpretable hybrid deep feature fusion network (IHDFN) as a solution to these problems. In the hybrid deep feature extraction module for protein sequences, shallow and deep features of protein sequences are extracted through two distinct views respectively, which capture multi-level information of proteins comprehensively. To further enhance the feature fusion effect, we introduce the StarNet fusion model in this module, enabling efficient fusion of shallow and deep features and enriching feature representation. To further improve the representation power of drug characteristics and the stability of the model, we use a graph convolutional network (GCN) in the drug feature extraction section in conjunction with residual connections and layer normalization. Furthermore, by integrating multimodal features from drugs and proteins utilizing an attention mechanism in the heterogeneous feature fusion module, we increase the complexity of features and achieve interpretability in predictions by attention focusing. Finally, we experimented on three datasets, and the findings indicate that IHDFN has exceptional performance and robustness compared to other cutting-edge techniques, underscoring its great promise and usefulness in DTI tasks. The code for this study is available on GitHub at https://github.com/wangqhfff/IHDFN.git .

Photoacoustic image reconstruction with the FD-UGAN model

Photoacoustic imaging has been widely used in the biomedical field by virtue of its high resolution and depth imaging advantages, but there are still problems with detector viewing angle limitation and incomplete data due to sparse sampling in the application process. To address this problem, FD-UGAN is proposed as a novel deep learning network, to our knowledge, in which the Full Dense U-Net is initially employed as the generator in a generative adversarial network framework for photoacoustic image reconstruction. The encoding and decoding pathways are redesigned to enable the network to efficiently capture and integrate multi-scale feature information, facilitating high-quality image reconstruction. The experimental results demonstrate that FD-UGAN outperforms existing reconstruction methods on the dataset. The proposed FD-UGAN provides an effective solution for image reconstruction from sparsely sampled data and exhibits significant potential for broader applications.

IGEV++: Iterative Multi-range Geometry Encoding Volumes for Stereo Matching.

Stereo matching is a core component in many computer vision and robotics systems. Despite significant advances over the last decade, handling matching ambiguities in ill-posed regions and large disparities remains an open challenge. In this paper, we propose a new deep network architecture, called IGEV++, for stereo matching. The proposed IGEV++ constructs Multi-range Geometry Encoding Volumes (MGEV), which encode coarse-grained geometry information for ill-posed regions and large disparities, while preserving fine-grained geometry information for details and small disparities. To construct MGEV, we introduce an adaptive patch matching module that efficiently and effectively computes matching costs for large disparity ranges and/or ill-posed regions. We further propose a selective geometry feature fusion module to adaptively fuse multi-range and multi-granularity geometry features in MGEV. Then, we input the fused geometry features into ConvGRUs to iteratively update the disparity map. MGEV allows to efficiently handle large disparities and ill-posed regions, such as occlusions and textureless regions, and enjoys rapid convergence during iterations. Our IGEV++ achieves the best performance on the Scene Flow test set across all disparity ranges, up to 768px. Our IGEV++ also achieves state-of-the-art accuracy on the Middlebury, ETH3D, KITTI 2012, and 2015 benchmarks. Specifically, IGEV++ achieves a 3.23% 2-pixel outlier rate (Bad 2.0) on the large disparity benchmark, Middlebury, representing error reductions of 31.9% and 54.8% compared to RAFT-Stereo and GMStereo, respectively. We also present a real-time version of IGEV++ that achieves the best performance among all published real-time methods on the KITTI benchmarks. The code is publicly available at https://github.com/gangweix/IGEV and https://github.com/gangweix/IGEV-plusplus.

GAMMNet: Gating Multi-head Attention in a Multi-modal Deep Network for Sound Based Respiratory Disease Detection.

Respiratory diseases present significant challenges to global health due to their high morbidity and mortality rates. Traditional diagnostic methods, such as chest radiographs and blood tests, often lead to unnecessary costs and resource strain, as well as potential risks of cross-contamination during these procedures. In recent years, contactless sensing and intelligent technologies, particularly multi-modal sound-based deep learning methods, have emerged as promising solutions for the early detection of respiratory diseases. While these methods have shown encouraging results, the integration of multi-modal features has not been sufficiently explored, which limits the enhancement of diagnostic accuracy. To address this issue, we introduce GAMMNet, a novel multi-modal neural network designed to enhance the detection of respiratory diseases by leveraging multi-modal sound data collected from contactless recording devices. GAMMNet utilizes a unique gating mechanism that adaptively regulates the influence of each modality on the classification results. Additionally, our model incorporates multi-head attention and linear transformation modules to further enhance classification performance. Our GAMMNet achieves state-of-the-art classification results, compared to existing deep learning based methods, on real-world multi-modal respiratory sound datasets. These findings demonstrate the robustness and effectiveness of GAMMNet in the contactless monitoring and early detection of respiratory diseases.

LCFANet: A Novel Lightweight Cross-Level Feature Aggregation Network for Small Agricultural Pest Detection

In agricultural pest detection, the small size of pests poses a critical hurdle to detection accuracy. To mitigate this concern, we propose a Lightweight Cross-Level Feature Aggregation Network (LCFANet), which comprises three key components: a deep feature extraction network, a deep feature fusion network, and a multi-scale object detection head. Within the feature extraction and fusion networks, we introduce the Dual Temporal Feature Aggregation C3k2 (DTFA-C3k2) module, leveraging a spatiotemporal fusion mechanism to integrate multi-receptive field features while preserving fine-grained texture and structural details across scales. This significantly improves detection performance for objects with large scale variations. Additionally, we propose the Aggregated Downsampling Convolution (ADown-Conv) module, a dual-path compression unit that enhances feature representation while efficiently reducing spatial dimensions. For feature fusion, we design a Cross-Level Hierarchical Feature Pyramid (CLHFP), which employs bidirectional integration—backward pyramid construction for deep-to-shallow fusion and forward pyramid construction for feature refinement. The detection head incorporates a Multi-Scale Adaptive Spatial Fusion (MSASF) module, adaptively fusing features at specific scales to improve accuracy for varying-sized objects. Furthermore, we introduce the MPDINIoU loss function, combining InnerIoU and MPDIoU to optimize bounding box regression. The LCFANet-n model has 2.78M parameters and a computational cost of 6.7 GFLOPs, enabling lightweight deployment. Extensive experiments on the public dataset demonstrate that the LCFANet-n model achieves a precision of 71.7%, recall of 68.5%, mAP50 of 70.4%, and mAP50-95 of 45.1%, reaching state-of-the-art (SOTA) performance in small-sized pest detection while maintaining a lightweight architecture.

Blind Signal Separation with Deep Residual Networks for Robust Synthetic Aperture Radar Signal Processing in Interference Electromagnetic Environments

With the rapid development of electronic technology, the electromagnetic interference encountered by airborne synthetic aperture radar (SAR) is no longer satisfied with a single type of interference, and it often encounters both suppressive and deceptive interference. In this manuscript, an algorithm based on blind signal separation (BSS) and deep residual learning is proposed for airborne SAR multi-electromagnetic interference suppression. Firstly, theoretical airborne SAR imaging in a multi-electromagnetic interference environment model is established, and the signal-mixed model of multi-electromagnetic interference is proposed. Then, a BSS algorithm using maximum kurtosis deconvolution and improved principal component analysis (PCA) is presented for suppressing the composite electromagnetic interference encountered by airborne SAR. Finally, in order to find the desired signal among multiple separated sources and to cope with the residual noise, a deep residual network is designed for signal recognition and denoising. This method uses a BSS algorithm with maximum kurtosis deconvolution and improved PCA to perform mixed signal separation. After performing signal separation, the original echo signal and the jamming can be obtained. To solve the separation order uncertainty and residual noise problems of the existing BSS algorithms, the deep residual network is designed to recognize airborne SAR signals after airborne SAR imaging. This algorithm has a better signal restoration degree, higher image restoration degree, and better compound interference suppression performance before and after anti-interference. Simulation and measurement results demonstrate the effectiveness of our presented algorithm.

Chaotic Time Series Forecasting by using Echo State Network and Autoregressive Model

Chaotic time series forecasting such as maximum wind speed rates is of great importance in the fields of meteorology and renewable energy to reduce and control the harmful negative effects. The problem of wind speed is that it is affected by several interrelated factors such as temperature and atmospheric pressure, which are characterized by non-linearity through the influence of time series on differences that may be a cause of the emergence of uncertainty problems, which makes it difficult to model using traditional univariate time series methods. Echo State Network (ESN) is a neural network specialized in time series forecasting after addressing the problem of synchronization with the time variable as a recurrent network to address time-dependent effects and accurate prediction of time series in addition to its ability to model nonlinearly. This study presents the use of the Autoregressive (AR) model and then its hybridization with the deep echo state network, which is called the AR-ESN hybrid method by using the optimal structure of the AR model to determine the optimal inputs to the ESN network as the main contributions to solving the prediction problems for real data forecasts. The use of ESN as a proposed forecasting method is to improve the forecasting efficiency to reduce the risks associated with extreme weather fluctuations compared with traditional forecasting results. The results indicate that the ESN model based on AR model can contribute to increasing the forecasting accuracy of maximum wind speed compared with traditional models by using mean absolute percentage error (MAPE) as one of the criteria the forecasting accuracy.

A new deep learning-based fast transcoding for internet of things applications

To achieve low-power video communication in Internet of Things, this study presents a new deep learning-based fast transcoding algorithm from distributed video coding (DVC) to high efficiency video coding (HEVC). The proposed method accelerates transcoding by minimizing HEVC encoding complexity. Specifically, it models the selections of coding unit (CU) partitions and prediction unit (PU) partition modes as classification tasks. To address these tasks, a novel lightweight deep learning network has been developed acting as the classifier in a top-down transcoding strategy for improved efficiency. The proposed transcoding algorithm operates efficiently at both CU and PU levels. At the CU level, it reduces HEVC encoding complexity by accurately predicting CU partitions. At the PU level, predicting PU partition modes for non-split CUs further streamlines the encoding process. Experimental results demonstrate that the proposed CU-level transcoding reduces complexity overhead by 45.69%, with a 1.33% average Bjøntegaard delta bit-rate (BD-BR) increase. At the PU level, the transcoding achieves an even greater complexity reduction, averaging 60.97%, with a 2.16% average BD-BR increase. These results highlight the algorithm’s efficiency in balancing computational cost and compression performance. The proposed method provides a promising low-power video coding scheme for resource-constrained terminals in both upstream and downstream video communication scenarios.

Remote Sensing Image Road Segmentation Based on Conditions Perceived 3D UX-Net

Abstract The rapid development of remote sensing technology has broadened channels for people to access information, making information extraction from remote sensing images an essential means of acquiring data. However, factors such as shadows, occlusions and spectral similarities between roads and other objects contribute to high mis-segmentation rates in existing deep learning semantic segmentation networks when segmenting roads in remote sensing images. Based on this, this paper proposes an improved version of the DDSA(Data-associated Deep Supervision Attention) network based on the three-dimensional user experience module for road segmentation. Simultaneously, an attention mechanism is integrated into the network to better focus on the crucial features of the input data and capture regions of interest. Finally, by adding deeper levels of supervision signals to different levels of the neural network, the network pays attention to different features of the input data, guiding the network to learn more discriminative features at each level, which effectively addresses the issue of poor performance in predicting unknown data. Experimental results demonstrate that, the DDSA UX Net(User Experience Network) exhibits outstanding performance in various metric scores and segmentation quality. These results support the potential of the DDSA UX Net network in applications such as urban development monitoring, traffic infrastructure planning and disaster management.