Loading...
主 管:江西省科学技术厅
主 办:江西省计算机学会
江西省计算中心
编辑出版:《计算机与现代化》编辑部
Toggle navigation
Home
About Journal
Editorial Board
Journal Online
Online First
Current Issue
Archive
Most Read Articles
Most Download Articles
Most Cited Articles
Subscription
FAQ
Self Recommendation
Contact Us
Email Alert
中文
Office Online
Submission or Manuscript
Peer Review
Editor-in-Chief
Office Work
Instruction
More>>
Instruction
Submit Flow
Template
Copyright Agreement
The Author Changes the Application
Highlights
More>>
Links
More>>
Visited
Total visitors:
Visitors of today:
Now online:
Table of Content
27 September 2024, Volume 0 Issue 09
Previous Issue
Collaborative Recommendation Algorithm with Implicit Roles
YU Tianyi, LI Jianfeng, CHEN Hailong, ZHAI Jun
2024, 0(09): 1-7. doi:
10.3969/j.issn.1006-2475.2024.09.001
Asbtract
(
125
)
PDF
(1594KB) (
121
)
References
|
Related Articles
|
Metrics
This article aims to improve the effectiveness of the algorithm, starts from the psychological needs of users, locates the implicit role group of users, and researches the personalized recommendation algorithms. From a theoretical point of view, the research in this paper effectively ensures the diversity requirements of recommendation systems and improves the accuracy of algorithms to a certain extent. It expands the relevant theory of implicit preference to address the phenomenon of preference evolution. Through verification in real data, multiple experimental evaluation indicators have been significantly improved. This not only provides a theoretical basis and reference for recommendation systems, but also improves the accuracy of recommendation results. It has broad application prospects. From a practical point of view, the classification of users in this article is no longer limited to ordinary social attributes, but can further explore users’ psychological needs, obtains more accurate and diverse recommendation results, improves user satisfaction and experience. Enterprises can guide users to change their interests, increase their loyalty and value, improve their lifecycle, and increase their profits.
Enhanced Big Language Model Dual Carbon Domain Services Based on Knowledge Graph
QI Jun1, 2, QU Ruiting2, JIAO Chuanming2, ZHOU Qiaoni2, GUO Yanliang3, TAN Wenjun3
2024, 0(09): 8-14. doi:
10.3969/j.issn.1006-2475.2024.09.002
Asbtract
(
136
)
PDF
(1796KB) (
124
)
References
|
Related Articles
|
Metrics
With the continuous development of the large language model, it has been widely applied in many fields. Due to the lack of knowledge in the dual carbon field in the big language model, the accuracy of the response results is low if the large language model is directly applied to the field of double carbon. Therefore, the method of constructing dual carbon knowledge graph as a knowledge base is adopted to enhance the application of large language models in the field of carbon peaking and carbon neutrality. The LoRA method is used to fine-tune the large language model to improve its ability to extract keywords in the carbon peaking and carbon neutrality fields. A dual carbon knowledge graph is constructed as local knowledge base to provide dual carbon domain knowledge for the model. The knowledge is used as the context of the problem, allowing the large language model to learn, and a prompt engineering assistance model is designed to generate responses. Finally, the effectiveness of the responses is evaluated. The experimental results show that, compared with the direct use of large language model, the method based on knowledge graph to enhance the dual carbon domain service of large language model has a high accuracy of intelligent response results in the field of carbon peaking and carbon neutrality, and provides an effective assistance for the construction of carbon peaking and carbon neutrality.
Dynamic Analysis Model of Reservoir Production Based on Improved
#br#
Time-series Capsule Network
ZHANG Huinan1, ZHANG Qiang1, SUN Hongxia2
2024, 0(09): 15-19. doi:
10.3969/j.issn.1006-2475.2024.09.003
Asbtract
(
76
)
PDF
(1233KB) (
64
)
References
|
Related Articles
|
Metrics
Many of China’s main development blocks in oilfields have gradually entered a high water cut period, with complex underground oil reservoirs and progressively increasing water content, leading to a decline in oil production. Improving the accurate understanding of the current stage of oilfield development and production patterns is of significant importance for studying the dynamic changes in oilfield production and formulating oilfield development strategies. In view of the dynamic change law of oilfield production, this paper proposes a dynamic analysis model for reservoirs based on improved time-series capsule prediction method. Firstly, a bidirectional gated recurrent unit is applied to capture the timing features in the oilfield data to enhance the fitting ability of the model to timing information. Secondly, the primary temporal feature information is captured with a multi-headed attentional deep convolutional layer to efficiently extract long-range dependencies and complex feature representations of sequences. Finally, in the dynamic routing algorithm, attention mechanism is introduced to allow the higher-level capsules to better focus on important features, so as to improve the efficiency and accuracy of information transmission. To verify the validity of the model, the time-series data of the oil field is used as input to predict the daily oil production by improving the output of the capsule network model. The improved capsule network is compared with nine models such as ResNet, LeNet. The experimental results show that the improved capsule network has higher prediction accuracy, it can reach 94.5%.
Speech Enhancement Based on Time-frequency Self-attention Residual Temporal
#br#
Convolutional Networks
HOU Congying, YANG Wengqing, WANG Zhao, CHENG Cong
2024, 0(09): 20-24. doi:
10.3969/j.issn.1006-2475.2024.09.004
Asbtract
(
84
)
PDF
(1564KB) (
89
)
References
|
Related Articles
|
Metrics
The main purpose of speech enhancement(SE) is to remove irrelevant signals such as noise. It is the front-end processing part of many speech processing tasks. SE plays an important role in fields such as video conferencing and live broadcasting. However, most studies on SE mainly focuses on the long-term context-dependent modeling of speech frames, without considering the energy distribution characteristics in the time-frequency domain. This paper proposes a self-attention module based on time-frequency domain, which makes it possible to explicitly introduce a priori thinking about speech distribution characteristics in the process of model modeling. Combined with the residual temporal convolutional network, a residual temporal convolutional network model based on time-frequency domain self-attention is constructed. In order to verify the validity of the model, two training targets, IRM and PSM, which are commonly used in the field of SE, are used for experiments. The experimental results show that the model significantly improves the performance in terms of four objective evaluation metrics in SE and is consistently better than other baseline models.
Dung Beetle Optimization Algorithm Integrating Multiple Strategies for Take-out Order Distribution Route Optimization
YANG Yufeng1, 2, XIA Xiaoyun2, CHEN Zefeng3, LIAO Weizhi2, LI Jiwu2
2024, 0(09): 25-32. doi:
10.3969/j.issn.1006-2475.2024.09.005
Asbtract
(
78
)
PDF
(1534KB) (
84
)
References
|
Related Articles
|
Metrics
With the development of takeaway industry, takeaway platforms have to select and plan efficient delivery routes among a large number of candidate orders. In order to satisfy the needs of both takeaway platforms and customers, an optimization model is established with the objectives of minimizing the delivery cost and maximizing the customer’s time satisfaction. The dung beetle optimization algorithm is used for solving the problem. Aiming at the problems that dung beetle optimization algorithm is prone to local optimization and low solving quality, a multi-strategy enhanced dung beetle optimization algorithm, named IDBO (Improved Dung Beetle Optimizer)is proposed by introducing simulated polynomial mutation strategy, simulated annealing probability jump operator and simplex local search strategy. The solution results obtained on randomly generated test cases show that the IDBO algorithm achieves better optimal solution, mean, standard deviation, cost and satisfaction compared to other algorithms. The simulation results show that the three improvement strategies can improve the optimization ability of the algorithm and solve the model effectively.
Automated Essay Scoring Method Based on GCN and Fine Tuned BERT
MA Yu, YANG Yong, REN Ge, Palidan Tuerxun
2024, 0(09): 33-37. doi:
10.3969/j.issn.1006-2475.2024.09.006
Asbtract
(
76
)
PDF
(1152KB) (
74
)
References
|
Related Articles
|
Metrics
Automatic scoring of essays is one of the important research directions in the field of smart education. It has the advantages of improving scoring efficiency, reducing labor costs, and ensuring the objectivity and consistency of scoring, so it has broad application prospects in the field of education. Although syntactic features play an important role in automatic scoring of compositions, there is still a lack of research on how to better utilize these features for automatic scoring of compositions. This paper proposes an automatic essay scoring method GFTB based on GCN and fine-tuned BERT. This model uses graph convolutional network to extract syntactic features of compositions, uses BERT and Adapter training methods to extract deep semantic features of compositions, and uses a gating mechanism to further capture the semantic features after the fusion of the two. The experimental results show that the proposed GFTB model achieves good average performance on 8 subsets of the public data set ASAP. Compared with baseline models such as Tongyi Qianwen, the proposed method can effectively improve the performance of automatic essay scoring.
Influenza-like Illness Prediction Based on LSTM-SIR-EAKF
LI Jin1, WEI Yanlong1, XUE Hongxin2, LIANG Haijian2
2024, 0(09): 38-44. doi:
10.3969/j.issn.1006-2475.2024.09.007
Asbtract
(
94
)
PDF
(1856KB) (
83
)
References
|
Related Articles
|
Metrics
The paper explores the combination method based on machine learning model and infectious disease model to predict influenza trend, and provides advice for medical institutions to take preventive measures. To precisely capture the temporal features of influenza-like illness (ILI), this paper proposes a combined prediction model (LSTM-SIR-EAKF) based on long and short-term memory(LSTM)neural networks, Suceptible-Infected-Recovered(SIR)model, and Ensemble Adjustment Kalman Filter(EAKF). Firstly, the model of LSTM is employed to learn the temporal relationship between ILI. Then, SIR model is used to simulate the transmission process of ILI. Finally, EAKF correctes the anticipated values of ILI from SIR model to obtain the final prediction values of ILI. The experimental results show that through the prediction of ILI in three time periods, the goodness of fit(R2)proposed by the LSTM-SIR-EAKF model are 0.996, 0.991 and 0.995, respectively, and the evaluation indicators of the prediction results are better than the comparison model. LSTM-SIR-EAKF model makes long-term prediction of influenza in time through long and short term memory network, and the infectious disease model simulates the changes of influenza population in space, effectively improving the prediction effect.
Cryptographic Algorithm of IoV Communication Based on AES
XU Xiaowei, CHENG Yu, QIAN Feng, ZHU Neng, DENG Mingxing
2024, 0(09): 45-51. doi:
10.3969/j.issn.1006-2475.2024.09.008
Asbtract
(
96
)
PDF
(2634KB) (
71
)
References
|
Related Articles
|
Metrics
As V2X technology develops rapidly, the volume of communication between vehicles and other devices, as well as the importance of information are growing rapidly, and the risk of in-vehicle information being attacked, intercepted or leaked has also increased accordingly, so the security of information interaction has become an unavoidable topic. Addressing the issues of large data volume and frequent data encryption and decryption operations in vehicle networking, this paper analyzes classical encryption algorithms and improves the traditional AES-based encryption algorithm. By using the RC4 encryption algorithm to generate a pseudo-random key instead of the key generation module of the AES encryption algorithm, the encryption time is optimized, and security performance is enhanced. Experiments are conducted to verify encryption efficiency and security.
Smart Delivery Service of Public Libraries Based on MTSP Problem
JIANG Xinzi1, AN Xiaoli1, GAO Shang2
2024, 0(09): 52-55. doi:
10.3969/j.issn.1006-2475.2024.09.009
Asbtract
(
50
)
PDF
(1180KB) (
51
)
References
|
Related Articles
|
Metrics
With the development of “Internet plus” thinking and library service mode and level, the logistics distribution of paper resources has become the last link of the library borrowing platform. How to reduce the delivery cost of the library, balance the workload of delivery personnel, and improve delivery efficiency in the smart library intelligent service platform is the research direction of smart services. In intelligent computing research, the ant colony algorithm is commonly used to solve the TSP traveling salesman problem, as it can utilize positive feedback and heuristic information induction to find the optimal solution for multi-objective traversal. Aiming at the multi-travel agent MTSP problem of interlibrary and community logistics distribution, the hybrid ant colony optimization algorithm is used to optimize the final distribution path of book paper resources, which can better realize the comprehensive improvement of delivery efficiency. The efficient and high-quality library services can better improve the quality of reading.
Text Clustering Method for Fragmented Reply Based on Dissimilarity Matrix
LIU Wenliang1, WU Fei1, HE Deming1, ZHAO Weiwei2, PAN Jianhong3
2024, 0(09): 56-60. doi:
10.3969/j.issn.1006-2475.2024.09.010
Asbtract
(
66
)
PDF
(1589KB) (
50
)
References
|
Related Articles
|
Metrics
In response to the problem of effectively extracting the required text information from fragmented reply texts in Q&A communities, this paper proposes a clustering method for fragmented reply texts based on dissimilarity matrix. Firstly, the clustering center is designed based on dissimilarity between texts and the fragmented reply texts in the community are classified by the clustering way. Then, the text features of user questions are extracted based on RNN+CNN. Finally, the automatic extraction of fragmented response text is achieved based on TF-IDF algorithm using the extracted question text features. The experimental results show that the proposed method can automatically extract the required text information with high accuracy and stability, and can be applied to the extraction of fragmented reply texts in question answering communities.
Survey of Digital Twin Modeling and Applications in Power System
LIU Ruoying1, ZOU Weiyu1, HU Shaoqian2, JI Shunhui3
2024, 0(09): 61-68. doi:
10.3969/j.issn.1006-2475.2024.09.011
Asbtract
(
92
)
PDF
(1236KB) (
65
)
References
|
Related Articles
|
Metrics
The power system is closely related to the production activities of various industries and people’s life. The digital twin technology can be used to effectively monitor the operation of the power system, respond in time, and reduce unnecessary time and labor costs. Based on the introduction of the concepts of power system and digital twin, this paper summarizes the research on the modeling and application of digital twin in power system in recent years. A systematic review is conducted on the relevant achievments of digital twin modeling in power systems from five perspectives: geometry, physics, behavior, rule, and multi-scale. The application of power system based on digital twin is summarized from five perspectives: fault detection, fault diagnosis, scheduling, state evaluation and multi-purpose. Finally, the challenges of digital twin modeling and application in power system are summarized, and the future development direction is explored.
Improved Pelican Optimization Algorithm Based on Circle Mapping and
#br#
Adaptive t-Distribution Mutation
GAO Meng, ZENG Xianwen
2024, 0(09): 69-73. doi:
10.3969/j.issn.1006-2475.2024.09.012
Asbtract
(
72
)
PDF
(2355KB) (
59
)
References
|
Related Articles
|
Metrics
In view of the shortcomings of the traditional pelican optimization algorithm, such as slow convergence speed and easy to fall into local optimal solutions, an improved pelican optimization algorithm based on Circle map initialization and adaptive t-distribution mutation is proposed. First, in the population initialization stage, the Circle mapping is used to generate an initial solution with a high degree of diversity, and combined with the reverse learning strategy, the diversity of the population is improved and the exploration ability of the population is enhanced. Secondly, in the iterative process, the adaptive t-distribution mutation operation is used to perturb the individual, which helps the pelican optimization algorithm jump out of the local optimal solution and improve the convergence speed. In addition, an adaptive factor and an improved inertia weight are introduced in the exploration stage of the pelican optimization algorithm, which better balances the global exploration ability and local development ability of the algorithm. Finally, IPOA is compared with other four classical algorithms on several test functions. Experimental results show that IPOA has a significant improvement in convergence speed, global search ability and convergence robustness.
Water Supply Pipeline Burr Data Detection Based on Improved LightGBM by Focal Loss
XUE Hao, MA Jing, GUO Xiaoyu
2024, 0(09): 74-81. doi:
10.3969/j.issn.1006-2475.2024.09.013
Asbtract
(
67
)
PDF
(1733KB) (
53
)
References
|
Related Articles
|
Metrics
Addressing the issue of low recall in the detection of burrs in water supply pipelines due to data imbalance, this paper proposes an improved method for detecting pipeline burr data by utilizing the Focal Loss function and integrating it with LightGBM. Firstly, considering the characteristics of pipeline burr data, neighborhood-related features are constructed. Secondly, the Focal Loss function is introduced into LightGBM to increase the model’s weight on hard-to-detect burr samples. Different parameter values for Focal Loss are experimented to balance precision and recall. Finally, different parameter settings for Focal Loss are selected for model fusion to further improve the detection performance of the model on imbalanced burr data. Experiments are carried out on a real dataset from a municipal water supply pipeline. The experimental results show that, compared with a single model based on the cross-entropy loss function, the fused model with the improved Focal Loss in this paper achieves 33.3 percentage points increase in recall and 18 percentage points increase in F1 score for burr data. However, the precision of burr data detection still needs further improvement. The method proposed in this paper starts with loss function and dynamically adjusts the weights of difficult and easy samples to effectively improve the detection performance of burr data under unbalanced data.
Public Opinion Guidance of Social Bots in Coupling Network Environment
Considering Temporal Characteristics
ZHANG Yaozeng, MA Jing
2024, 0(09): 82-90. doi:
10.3969/j.issn.1006-2475.2024.09.014
Asbtract
(
73
)
PDF
(4393KB) (
50
)
References
|
Related Articles
|
Metrics
In recent years, social bots have garnered significant attention as new participants in the ecosystem of social networks. However, research on them within the country is relatively scarce. To gain in-depth insights into the role and corresponding patterns of social bots in the evolution of public opinion, it is necessary to investigate their impact on the evolution of group opinions. This study, by considering the Hawkes process manifested in individual activity changes, constructs a time-varying driving mechanism for the formation of social network connections. Simultaneously, it establishes an asymmetric two-layer coupled social network and the corresponding continuous opinion evolution model. The study introduces social bots and explores their guiding effects on group opinions. The experimental analysis reveals that the discussion in online social networks about public events heats up and triggers a further increase in the group activity level in offline social networks, aligning with real-world scenarios. Additionally, a very small number of social bots can effectively guide group opinions, demonstrating their practical value. Therefore, setting a certain number of social bots and deploying them early on is an effective strategy for guiding public opinion.
Named Entity Recognition in Field of Party Building Based on BERT-BiLSTM-CRF
ZHAO Dun1, SHE Xuebing2, WU Changxing3
2024, 0(09): 91-94. doi:
10.3969/j.issn.1006-2475.2024.09.015
Asbtract
(
67
)
PDF
(1771KB) (
49
)
References
|
Related Articles
|
Metrics
When constructing a knowledge graph in the field of party building, the traditional named entity recognition (NER) methods often suffer from unclear entity boundaries and polysemy of entity terms, which lead to low recognition accuracy and efficiency. To address these issues, this paper proposes a BERT-BiLSTM-CRF entity recognition model that integrates tree-like probability and a domain dictionary. The model involves embedding the domain dictionary into BERT for text vectorization, utilizes BiLSTM to acquire contextual semantic features, and applies tree-like probability to the transition probability calculation in the CRF layer to enhance word segmentation accuracy. The experimental results on the MSRA and self-constructed corpora, compared with the baseline model, show that the proposed model achieves better performance in terms of F1-score, recall, and precision.
Boundary Mixed Resampling Based on Joint Entropy for Imbalanced Data
ZHOU Chuanhua1, 2, REN Taijiao1, LUO Lan1, ZHOU Hao1
2024, 0(09): 95-100. doi:
10.3969/j.issn.1006-2475.2024.09.016
Asbtract
(
62
)
PDF
(1197KB) (
51
)
References
|
Related Articles
|
Metrics
In order to overcome the limitations of single resampling methods in data imbalance handling, which often lead to the generation of redundant samples and the inadvertent deletion of crucial sample information, this paper proposes a novel non-balanced data boundary mixed resampling algorithm based on joint entropy. The algorithm first effectively distinguishes between the boundary set and the non-boundary set by introducing a boundary factor. It further constructs a joint entropy indicator system to assess the importance of minority class samples within the boundary set. Based on this assessment, different oversampling methods and sampling quantities are applied to the segmented minority class samples. Finally, the NearMiss-2 algorithm is used to filter and remove most of the sample points in the non-boundary set, thus achieving a relative data balance. Through comparative experiments on nine sets of UCI datasets, the experimental results show that the proposed algorithm achieves improvements in F1-Score, G-mean, and AUC metrics, which validates its effectiveness and exhibiting favorable performance in non-balanced data classification.
Short Text Classification Combining Attention Mechanism and Mengzi Model
CHEN Xuesong1, LI Heng1, WANG Haochang2
2024, 0(09): 101-106. doi:
10.3969/j.issn.1006-2475.2024.09.017
Asbtract
(
68
)
PDF
(1256KB) (
68
)
References
|
Related Articles
|
Metrics
How to use short text classification technology to mine useful text information is one of the current hot research directions. To solve the problem of sparse feature information and difficult extraction of short text, a short text classification model named Mengzi-ADCBU is proposed. This model uses Mengzi pre-training model to convert input text information into corresponding text representation. Then, the obtained text vectors are input to the improved deep pyramid convolutional neural network and the bidirectional gated unit integrated with multi-head attention mechanism to extract text feature information, and the extracted feature information is fused and sent to the full connection layer and Softmax function to complete short text classification. Multiple models comparison experiments are carried out on the publicly available THUCNews short text data set and SougouCS short text data set respectively. The experimental results show that the proposed Mengzi-ADCBU model is better than the current mainstream models in the accuracy, precision, recall rate and F1 value of short text classification and has better short text classification ability.
Camera Module Defect Detection Based on Improved YOLOv8s
ZHANG Ze1, ZHANG Jianquan2, 3, ZHOU Guopeng2, 3
2024, 0(09): 107-113. doi:
10.3969/j.issn.1006-2475.2024.09.018
Asbtract
(
90
)
PDF
(3880KB) (
92
)
References
|
Related Articles
|
Metrics
Aiming at the problems of the great change of defect size, unclear contour and high missed detection rate of small target defects in camera module defect detection, an improved YOLOv8s algorithm is proposed. Firstly, the small target detection layer is added to improve the detection performance of small targets. Secondly, BiFormer is introduced to improve the C2f module in the backbone network, and the C2f-Bif module is proposed to enhance the ability of the network to extract image features. Then, the H-SPPF (Hybrid Fast Space Pyramid Pooling) module is proposed to enhance the ability of the network to capture local and global information. Finally, the parameter-free SimAM attention mechanism is added to suppress the non-target background interference information and improve the attention of the target. The experimental results show that the average accuracy of the improved YOLOv8s algorithm for camera module defect detection reaches 87.2% under the condition of reducing the number of model parameters, which is 3.2 percentage points higher than that of the YOLOv8s algorithm. The detection speed reaches 55 FPS, which meets the factory’s real-time detection requirements for camera module defects.
Recognition and Warning of Elevator Abnormal Behavior Based on Human Skeleton
YU Chenxi, GU Lin
2024, 0(09): 114-120. doi:
10.3969/j.issn.1006-2475.2024.09.019
Asbtract
(
68
)
PDF
(4225KB) (
67
)
References
|
Related Articles
|
Metrics
In order to accurately identify passengers’ abnormal behaviors such as fighting in the closed and narrow elevator car, and avoid the occurrence of safety accidents, a passenger abnormal behavior identification method based on the joint spatio-temporal features of human skeleton is proposed. Firstly, this method uses YOLOv7 to detect the passenger’s position in the video, extracts the coordinates of the key points of the skeleton through the YOLOv7-Pose pose estimation algorithm, and filter out the complex background interference. Secondly, for the features of large amplitude, fast speed, and chaotic direction of the abnormal behaviors, we use the SURF joint pyramid hierarchical improvement of the LK optical flow method to carry out joint temporal and spatial feature extraction of the passenger’s human skeleton information. Finally, the optical flow changes of the feature points can be used to judge whether abnormal behavior occurs in the car and alarm it in time. The dataset in this paper is derived from the self-constructed dataset in elevator scenario and the behavioral public dataset in non-elevator scenario respectively. After experimental validation, the accuracy of this method on the recognition of abnormal behaviors reaches 95.53%, which is improved in speed and accuracy compared with other methods. It can meet the real-time requirements and be applied to the video monitoring system of elevator car to ensure the safety of the passengers in the elevator.
Multi-scale Depth Fusion Monocular Depth Estimation Based on Transposed Attention
CHENG Yazi1, LEI Liang1, 2, CHEN Han1, ZHAO Yiran1
2024, 0(09): 121-126. doi:
10.3969/j.issn.1006-2475.2024.09.020
Asbtract
(
68
)
PDF
(1162KB) (
69
)
References
|
Related Articles
|
Metrics
Monocular depth estimation is a fundamental task in computer vision, aiming to predict depth maps from single images and retrieve depth information for corresponding pixel positions. This paper proposes a novel network architecture for monocular depth estimation to further enhance the predictive accuracy of the network. Transposed attention introduces a self-attention mechanism, enabling it to focus on specific regions within the image while reducing the parameter and computation requirements. By incorporating information across different channels, it effectively captures fine-grained regions and edge details for learning. The paper presents an improved version of transposed attention that retains semantic information with fewer parameters. Multi-scale depth fusion leverages the characteristic of extracting features with different depths from distinct channels. It computes the average depth for each channel, enhancing the model’s depth perception capability. Furthermore, it models long-range dependencies for vertical distances, effectively separating edges between objects and mitigating the loss of fine-grained information. Finally, the proposed modules’ effectiveness is validated through experiments conducted on the NYU Depth V2 dataset and the KITTI dataset, demonstrating exceptional performance.