Structural Feature Extraction Research Articles

Integrating Prior Chemical Knowledge into the Graph Transformer Network to Predict the Stability Constants of Chelating Agents and Metal Ions.

The latest advancements in nuclear medicine indicate that radioactive isotopes and associated metal chelators play crucial roles in the diagnosis and treatment of diseases. The development of metal chelators mainly relies on traditional trial-and-error methods, lacking rational guidance and design. In this study, we propose the structure-aware transformer (SAT) combined with molecular fingerprint (SATCMF), a novel graph transformer network framework that incorporates prior chemical knowledge to construct coordination edges and learns the interactions between chelating agents and metal ions. SATCMF is trained on stability data collected from metal ion-ligand complexes, leveraging the SAT network to extract structural features relevant to the binding of ligands with metal ions. It further integrates molecular fingerprint features to refine the prediction of the stability constants of the chelating agents and metal ions. The experimental results on benchmark data set demonstrate that SATCMF achieves state-of-the-art performance based on four different graph neural network architectures. Additionally, visualizing the learned molecular attention distribution provides interpretable insights from the prediction results, offering valuable guidance for the development of novel metal chelators.

A dataset of alternately located segments in protein crystal structures

Protein Data Bank (PDB) files list the relative spatial location of atoms in a protein structure as the final output of the process of fitting and refining to experimentally determined electron density measurements. Where experimental evidence exists for multiple conformations, atoms are modelled in alternate locations. Programs reading PDB files commonly ignore these alternate conformations by default leaving users oblivious to the presence of alternate conformations in the structures they analyze. This has led to underappreciation of their prevalence, under characterisation of their features and limited the accessibility to this high-resolution data representing structural ensembles. We have trawled PDB files to extract structural features of residues with alternately located atoms. The output includes the distance between alternate conformations and identifies the location of these segments within the protein chain and in proximity of all other atoms within a defined radius. This dataset should be of use in efforts to predict multiple structures from a single sequence and support studies investigating protein flexibility and the association with protein function.

Gastrointestinal tract disease detection via deep learning based structural and statistical features optimized hexa-classification model.

Gastrointestinal tract (GIT) diseases impact the entire digestive system, spanning from the mouth to the anus. Wireless Capsule Endoscopy (WCE) stands out as an effective analytic instrument for Gastrointestinal tract diseases. Nevertheless, accurately identifying various lesion features, such as irregular sizes, shapes, colors, and textures, remains challenging in this field. Several computer vision algorithms have been introduced to tackle these challenges, but many relied on handcrafted features, resulting in inaccuracies in various instances. In this work, a novel Deep SS-Hexa model is proposed which is a combination two different deep learning structures for extracting two different features from the WCE images to detect various GIT ailment. The gathered images are denoised by weighted median filter to remove the noisy distortions and augment the images for enhancing the training data. The structural and statistical (SS) feature extraction process is sectioned into two phases for the analysis of distinct regions of gastrointestinal. In the first stage, statistical features of the image are retrieved using MobileNet with the support of SiLU activation function to retrieve the relevant features. In the second phase, the segmented intestine images are transformed into structural features to learn the local information. These SS features are parallelly fused for selecting the best relevant features with walrus optimization algorithm. Finally, Deep belief network (DBN) is used classified the GIT diseases into hexa classes namely normal, ulcer, pylorus, cecum, esophagitis and polyps on the basis of the selected features. The proposed Deep SS-Hexa model attains an overall average accuracy of 99.16% in GIT disease detection based on KVASIR and KID datasets. The proposed Deep SS-Hexa model achieves high level of accuracy with minimal computational cost in the recognition of GIT illness. The proposed Deep SS-Hexa Model progresses the overall accuracy range of 0.04%, 0.80% better than GastroVision, Genetic algorithm based on KVASIR dataset and 0.60%, 1.21% better than Modified U-Net, WCENet based on KID dataset respectively.

Structural prior-driven feature extraction with gradient-momentum combined optimization for convolutional neural network image classification

Recent image classification efforts have achieved certain success by incorporating prior information such as labels and logical rules to learn discriminative features. However, these methods overlook the variability of features, resulting in feature inconsistency and fluctuations in model parameter updates, which further contribute to decreased image classification accuracy and model instability. To address this issue, this paper proposes a novel method combining structural prior-driven feature extraction with gradient-momentum (SPGM), from the perspectives of consistent feature learning and precise parameter updates, to enhance the accuracy and stability of image classification. Specifically, SPGM leverages a structural prior-driven feature extraction (SPFE) approach to calculate gradients of multi-level features and originalimages to construct structural information, which is then transformed into prior knowledge to drive the network to learn features consistent with the original images. Additionally, an optimization strategy integrating gradients and momentum (GMO) is introduced, dynamically adjusting the direction and step size of parameter updates based on the angle and norm of the sum of gradients and momentum, enabling precise model parameter updates. Extensive experiments on CIFAR10 and CIFAR100 datasets demonstrate that the SPGM method significantly reduces the top-1 error rate in image classification, enhances the classification performance, and outperforms state-of-the-art methods.

Structural prior-driven feature extraction with gradient-momentum combined optimization for convolutional neural network image classification

Recent image classification efforts have achieved certain success by incorporating prior information such as labels and logical rules to learn discriminative features. However, these methods overlook the variability of features, resulting in feature inconsistency and fluctuations in model parameter updates, which further contribute to decreased image classification accuracy and model instability. To address this issue, this paper proposes a novel method combining structural prior-driven feature extraction with gradient-momentum (SPGM), from the perspectives of consistent feature learning and precise parameter updates, to enhance the accuracy and stability of image classification. Specifically, SPGM leverages a structural prior-driven feature extraction (SPFE) approach to calculate gradients of multi-level features and originalimages to construct structural information, which is then transformed into prior knowledge to drive the network to learn features consistent with the original images. Additionally, an optimization strategy integrating gradients and momentum (GMO) is introduced, dynamically adjusting the direction and step size of parameter updates based on the angle and norm of the sum of gradients and momentum, enabling precise model parameter updates. Extensive experiments on CIFAR10 and CIFAR100 datasets demonstrate that the SPGM method significantly reduces the top-1 error rate in image classification, enhances the classification performance, and outperforms state-of-the-art methods.

Stacked deformable convolution network with weighted non-local attention and branch residual connection for image quality assessment

Convolutional neural networks are data-driven Image Quality Assessment (IQA) models based on the convolutional operation in a rectangular window. Deformable convolutions with learnable receptive fields can efficiently extract structural features of irregular objects in an image. By superimposing deformable convolutions, a plug-and-play module is designed to obtain information for irregular geometric shapes. To selectively fuse shallow and deep features, we propose a weighted non-local attention (WNLA) module with the input and output of self-attention in a weighted manner. This paper proposes a dual branch residual full-reference IQA network that combines weighted non-local attention and stacked deformable convolution. The proposed network was trained on PIPAL dataset and tested on LIVE and TID2013. The cross-dataset evaluation shows that the network has a competitive generalization ability. Ablation experiments indicate that the proposed modules can effectively improve the performance of the network. Comparative experiments reveal that our network is superior to existing excellent networks. The codes for training, test and visualization are available at: https://github.com/Pengchang-haha/SDCN.git.

Mitigating imbalances in heterogeneous feature fusion for multi-class 6D pose estimation

Most 6D pose studies often treat RGB and Depth features equally in fusion, potentially limiting model generalization, especially in multi-class tasks. This limitation arises from prevalent static map generation strategies that overlook discriminative features in downsampling sparse point clouds. Additionally, the commonly adopted direct concatenation approach in heterogeneous feature fusion often leads to an averaging effect, thereby reducing the effectiveness of each feature. To tackle these challenges, we propose an effective model for dynamic graph structure feature extraction, aimed at capturing richer features from point clouds. And we introduce an adaptive fusion method for heterogeneous features, which takes into account the unequal contributions to 6D pose estimation. Validation on benchmark datasets LineMOD and YCB-Video demonstrates its effectiveness for multi-class 6D pose estimation, surpassing existing fusion methods. Of particular significance, our method attains state-of-the-art (SOTA) results on the YCB-Video dataset. The code for this study can be accessed at https://github.com/ZEROhands/6D_Pose_Estimate.

A dual fusion recognition model for sleep posture based on air mattress pressure detection

In order to solve the difficult portability problem of traditional non-invasive sleeping posture recognition algorithms arising from the production cost and computational cost, this paper proposes a sleeping posture recognition model focusing on human body structural feature extraction and integration of feature space and algorithms based on a specific air-spring mattress structure, called SPR-DE (SPR-DE is the Sleep Posture Recognition-Data Ensemble acronym form). The model combines SMR (SMR stands for Principle of Spearman Maximal Relevance) with horizontal and vertical division based on the barometric pressure signals in the human body’s backbone region to reconstruct the raw pressure data into strongly correlated non-image features of the sleep postures in different parts and directions and construct the feature set. Finally, the recognit-ion of the two sleep postures is accomplished using the AdaBoost-SVM integrated classifier. SPR-DE is compared with the base and integrated classifiers to verify its performance. The experimental results show that the amount of significant features helps the algorithm to classify different sleeping patterns more accurately, and the f1 score of the SPR-DE model determined by the comparison experiments is 0.998, and the accuracy can reach 99.9%. Compared with other models, the accuracy is improved by 2.9% ~ 7.7%, and the f1-score is improved by 0.029 ~ 0.076. Therefore, it is concluded that the SMR feature extraction strategy in the SPR-DE model and the AdaBoost-SVM can achieve high accuracy and strong robustness in the task of sleep posture recognition in a small area, low-density air-pressure mattress, taking into account the comfort of the mattress structural design and the sleep posture recognition, integrated with the mattress adaptive adjustment system.

CKG-IMC: An inductive matrix completion method enhanced by CKG and GNN for Alzheimer’s disease compound-protein interactions prediction

Alzheimer’s disease (AD) is one of the most prevalent chronic neurodegenerative disorders globally, with a rapidly growing population of AD patients and currently no effective therapeutic interventions available. Consequently, the development of therapeutic anti-AD drugs and the identification of AD targets represent one of the most urgent tasks. In this study, in addition to considering known drugs and targets, we explore compound-protein interactions (CPIs) between compounds and proteins relevant to AD. We propose a deep learning model called CKG-IMC to predict Alzheimer’s disease compound-protein interaction relationships. CKG-IMC comprises three modules: a collaborative knowledge graph (CKG), a principal neighborhood aggregation graph neural network (PNA), and an inductive matrix completion (IMC). The collaborative knowledge graph is used to learn semantic associations between entities, PNA is employed to extract structural features of the relationship network, and IMC is utilized for CPIs prediction. Compared with a total of 16 baseline models based on similarities, knowledge graphs, and graph neural networks, our model achieves state-of-the-art performance in experiments of 10-fold cross-validation and independent test. Furthermore, we use CKG-IMC to predict compounds interacting with two confirmed AD targets, 42-amino-acid β-amyloid (Aβ42) protein and microtubule-associated protein tau (tau protein), as well as proteins interacting with five FDA-approved anti-AD drugs. The results indicate that the majority of predictions are supported by literature, and molecular docking experiments demonstrate a strong affinity between the predicted compounds and targets.

The Attention-Based Autoencoder for Network Traffic Classification with Interpretable Feature Representation

Network traffic classification is crucial for identifying network applications and defending against network threats. Traditional traffic classification approaches struggle to extract structural features and suffer from poor interpretability of feature representations. The high symmetry between network traffic classification and its interpretable feature representation is vital for network traffic analysis. To address these issues, this paper proposes a traffic classification and feature representation model named the attention mechanism autoencoder (AMAE). The AMAE model extracts the global spatial structural features of network traffic through attention mechanisms and employs an autoencoder to extract local structural features and perform dimensionality reduction. This process maps different network traffic features into one-dimensional coordinate systems in the form of spectra, termed FlowSpectrum. The spectra of different network traffic represent different intervals in the coordinate system. This paper tests the interpretability and classification performance of network traffic features of the AMAE model using the ISCX-VPN2016 dataset. Experimental results demonstrate that by analyzing the overall distribution of attention weights and local weight values of network traffic, the model effectively explains the differences in the spectral representation intervals of different types of network traffic. Furthermore, our approach achieves the highest classification accuracy of up to 100% for non-VPN-encrypted traffic and 99.69% for VPN-encrypted traffic, surpassing existing traffic classification schemes.

Accurate Prediction of Lysine Methylation Sites Using Evolutionary and Structural-Based Information

Methylation is considered one of the proteins’ most important post-translational modifications (PTM). Plasticity and cellular dynamics are among the many traits that are regulated by methylation. Currently, methylation sites are identified using experimental approaches. However, these methods are time-consuming and expensive. With the use of computer modelling, methylation sites can be identified quickly and accurately, providing valuable information for further trial and investigation. In this study, we propose a new machine-learning model called MeSEP to predict methylation sites that incorporates both evolutionary and structural-based information. To build this model, we first extract evolutionary and structural features from the PSSM and SPD2 profiles, respectively. We then employ Extreme Gradient Boosting (XGBoost) as the classification model to predict methylation sites. To address the issue of imbalanced data and bias towards negative samples, we use the SMOTETomek-based hybrid sampling method. The MeSEP was validated on an independent test set (ITS) and 10-fold cross-validation (TCV) using lysine methylation sites. The method achieved: an accuracy of 82.9% in ITS and 84.6% in TCV; precision of 0.92 in ITS and 0.94 in TCV; area under the curve values of 0.90 in ITS and 0.92 in TCV; F1 score of 0.81 in ITS and 0.83 in TCV; and MCC of 0.67 in ITS and 0.70 in TCV. MeSEP significantly outperformed previous studies found in the literature. MeSEP as a standalone toolkit and all its source codes are publicly available at https://github.com/arafatro/MeSEP.

Multi-resolution sequence and structure feature extraction for binding site prediction

Circular ribonucleic acids (circRNAs) are single-stranded RNA molecules that form loops and are widely expressed in various cells and tissues. They interact with RNA-binding proteins (RBPs) and play a vital regulatory role in the onset and development of several diseases. Researchers have proposed various hybrid architecture prediction methods based on convolutional neural networks and recurrent neural networks to recognize the interactions and sites between circRNAs and RBPs and thus reveal the biological functions of circRNAs. However, existing methods usually ignore the structural information of circRNA, which may affect the modeling of circRNA and RBP binding modes. To address these problems, we propose a prediction model based on multi-resolution feature extraction. First, it generates sequence features using unsupervised word embedding and nucleotide density. Then, it uses implicit and explicit pseudo-secondary structure hybrid encoding to fuse sequence and structural information and better simulate circRNA-RBP binding patterns. Second, it uses an enhanced bidirectional sample convolution and interaction network encoder to capture and integrate high-order features of distinct resolutions from the multi-scale convolution module. This provides rich semantic input to the downstream bidirectional long short-term memory network to improve prediction accuracy. Experimental results on 37 circRNA and 31 linear RNA datasets show that our method has significant advantages in identifying RNA-RBP interactions. Furthermore, the four motifs learned by our method are verified against existing motif databases, indicating that it can discover biologically meaningful circRNA-RBP binding patterns.

Gabor-net with multi-scale hierarchical fusion of features for fundus retinal blood vessel segmentation

This paper proposes a fundus retinal blood vessel segmentation model based on a deep convolutional network structure and biological visual feature extraction mechanism. It aims to solve the multi-scale problem of blood vessels in the fundus retinal blood vessel segmentation task in the field of medical image processing on the basis of increasing the biological interpretability of the model. First, the subject feature information of the retinal blood vessel image is obtained by using the non-subsampled Residual Bolck convolution main channel. Secondly, combined with the study of biological vision mechanisms, an information processing model of the Retina-Exogenius-Primary visual cortex (V1) ventral visual pathway was established. Gabor functions of different scales are used to simulate the structure of different levels of the visual pathway, and the scale information at different levels is integrated into the corresponding hierarchical stages of the convolutional main pathway network to enrich the information of small blood vessels and enhance the semantic information of the overall blood vessels. Finally, considering the imbalance of the ratio of vessel and nonvessel pixels, an adaptive optimization scheme using hybrid loss function weights is proposed to enhance the priority of blood vessel pixels in the calculation of the loss function. According to the experimental results on the STARE, DRIVE and CHASE_DB1 data sets, the model still achieves superior performance evaluation indicators overall compared with the existing optimal methods in the fundus retinal blood vessel segmentation task. This research is of great significance to the field of medical image processing and can provide more accurate auxiliary diagnostic information for clinical diagnosis and treatment.

Research on Visual Perception of Speed Bumps for Intelligent Connected Vehicles Based on Lightweight FPNet.

In the field of intelligent connected vehicles, the precise and real-time identification of speed bumps is critically important for the safety of autonomous driving. To address the issue that existing visual perception algorithms struggle to simultaneously maintain identification accuracy and real-time performance amidst image distortion and complex environmental conditions, this study proposes an enhanced lightweight neural network framework, YOLOv5-FPNet. This framework strengthens perception capabilities in two key phases: feature extraction and loss constraint. Firstly, FPNet, based on FasterNet and Dynamic Snake Convolution, is developed to adaptively extract structural features of distorted speed bumps with accuracy. Subsequently, the C3-SFC module is proposed to augment the adaptability of the neck and head components to distorted features. Furthermore, the SimAM attention mechanism is embedded within the backbone to enhance the ability of key feature extraction. Finally, an adaptive loss function, Inner-WiseIoU, based on a dynamic non-monotonic focusing mechanism, is designed to improve the generalization and fitting ability of bounding boxes. Experimental evaluations on a custom speed bumps dataset demonstrate the superior performance of FPNet, with significant improvements in key metrics such as the mAP, mAP50_95, and FPS by 38.76%, 143.15%, and 51.23%, respectively, compared to conventional lightweight neural networks. Ablation studies confirm the effectiveness of the proposed improvements. This research provides a fast and accurate speed bump detection solution for autonomous vehicles, offering theoretical insights for obstacle recognition in intelligent vehicle systems.

G-K BertDTA: A graph representation learning and semantic embedding-based framework for drug-target affinity prediction

Developing new drugs is costly, time-consuming, and risky. Drug-target affinity (DTA), indicating the binding capability between drugs and target proteins, is a crucial indicator for drug development. Accurately predicting interaction strength between new drug-target pairs by analyzing previous experiments aids in screening potential drug molecules, repurposing them, and developing safe and effective medicines. Existing computational models for DTA prediction rely on strings or single-graph neural networks, lacking consideration of protein structure and molecular semantic information, leading to limited accuracy. Our experiments demonstrate that string-based methods may overlook protein conformations, causing a high root mean square error (RMSE) of 3.584 in affinity due to a lack of spatial context. Single graph networks also underperform on topology features, with a 6% lower confidence interval (CI) for activity classification. Absent semantic information also limits generalization across diverse compounds, resulting in 18% increment in RMSE and 5% in misclassifications within quantifications study, restricting potential drug discovery. To address these limitations, we propose G-K BertDTA, a novel framework for accurate DTA prediction incorporating protein features, molecular semantic features, and molecular structural information. In this proposed model, we represent drugs as graphs, with a GIN employed to learn the molecular topological information. For the extraction of protein structural features, we utilize a DenseNet architecture. A knowledge-based BERT semantic model is incorporated to obtain rich pre-trained semantic embeddings, thereby enhancing the feature information. We extensively evaluated our proposed approach on the publicly available benchmark datasets (i.e., KIBA and Davis), and experimental results demonstrate the promising performance of our method, which consistently outperforms previous state-of-the-art approaches. Code is available at https://github.com/AmbitYuki/G-K-BertDTA.

MINDG: a drug-target interaction prediction method based on an integrated learning algorithm.

Drug-target interaction (DTI) prediction refers to the prediction of whether a given drug molecule will bind to a specific target and thus exert a targeted therapeutic effect. Although intelligent computational approaches for drug target prediction have received much attention and made many advances, they are still a challenging task that requires further research. The main challenges are manifested as follows: (i) most graph neural network-based methods only consider the information of the first-order neighboring nodes (drug and target) in the graph, without learning deeper and richer structural features from the higher-order neighboring nodes. (ii) Existing methods do not consider both the sequence and structural features of drugs and targets, and each method is independent of each other, and cannot combine the advantages of sequence and structural features to improve the interactive learning effect. To address the above challenges, a Multi-view Integrated learning Network that integrates Deep learning and Graph Learning (MINDG) is proposed in this study, which consists of the following parts: (i) a mixed deep network is used to extract sequence features of drugs and targets, (ii) a higher-order graph attention convolutional network is proposed to better extract and capture structural features, and (iii) a multi-view adaptive integrated decision module is used to improve and complement the initial prediction results of the above two networks to enhance the prediction performance. We evaluate MINDG on two dataset and show it improved DTI prediction performance compared to state-of-the-art baselines. https://github.com/jnuaipr/MINDG.

Holographic matrix method for rapid material structural feature extraction and detection

The extraction of the structural features of materials is fundamental for investigating novel properties in fields such as electronic information and biochemistry. However, existing experimental methods have limitations in analyzing material structures with sufficient depth. Therefore, rapid and accurate extraction and analysis of structural features from atomic coordinates obtained through simulation calculations are crucial for advancing the exploration of new material properties. Herein, we propose an approach for extracting the structural features of materials by combining the holographic matrix method with Bayesian optimization and tensor flow operations. The proposed algorithm efficiently classifies and statistically analyzes cluster structures within materials. Experimental validation conducted on a system comprising 8000 atoms demonstrated a correct recognition rate exceeding 99.213%. Moreover, the algorithm achieved an average recognition time of approximately [Formula: see text][Formula: see text]s. The proposed analytical framework exhibits scalability and robustness, establishing an algorithmic foundation for future advancements in big data analytics for complex materials.

Abstract PR016: Topology of the tumor microenvironment: Integrating imaging, modeling, and topological data analysis

Abstract Cancer evolution involves multiple stochastic processes across many scales. In addition to genomic and transcriptomic changes, hallmarks of cancer within solid tumors include the malignant co-evolution of the tumor microenvironment. Although the evaluation of tissue architecture plays a key role in pathologist staging, the automated extraction of additional structural features remains far from translation to the clinic. However, one particularly effective novel methodology for extracting patterns from these clinical images is topological data analysis(TDA). TDA is a field that studies the shape of data and its connectedness using concepts from geometric and algebraic topology. Recent work using topological data analysis has allowed the extraction of image features. Such features have been associated with clinical outcomes in both the analysis of H&E images in prostate cancer and CT images in lung cancer. We set out to question how the topological features of tissues, including tumor and stromal composition, evolve during treatment and between original sites and metastasis. To analyze this we use a combination of tissue sections, imaging, and cell-automata models and persistent homology and cubical persistence techniques from TDA. We apply certain topological methods for the first time to tissue images (H&E) and to the output of spatial simulations. We focus specifically on analyzing the evolution of topological features in ovarian and lung cancer and compare our features to 2D and 3D simulations (HAL) with varying game theoretical interactions as well as comparing features across tissues from matched primary and metastatic sites. We find that several topological features are conserved between primary and metastatic sites and some feature evolution is replicated in simulation. Citation Format: Rowan Barker-Clarke, Jason M. Gray, Jacob G. Scott. Topology of the tumor microenvironment: Integrating imaging, modeling, and topological data analysis [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Translating Cancer Evolution and Data Science: The Next Frontier; 2023 Dec 3-6; Boston, Massachusetts. Philadelphia (PA): AACR; Cancer Res 2024;84(3 Suppl_2):Abstract nr PR016.

Lactylation prediction models based on protein sequence and structural feature fusion.

Lysine lactylation (Kla) is a newly discovered posttranslational modification that is involved in important life activities, such as glycolysis-related cell function, macrophage polarization and nervous system regulation, and has received widespread attention due to the Warburg effect in tumor cells. In this work, we first design a natural language processing method to automatically extract the 3D structural features of Kla sites, avoiding potential biases caused by manually designed structural features. Then, we establish two Kla prediction frameworks, Attention-based feature fusion Kla model (ABFF-Kla) and EBFF-Kla, to integrate the sequence features and the structure features based on the attention layer and embedding layer, respectively. The results indicate that ABFF-Kla and Embedding-based feature fusion Kla model (EBFF-Kla), which fuse features from protein sequences and spatial structures, have better predictive performance than that of models that use only sequence features. Our work provides an approach for the automatic extraction of protein structural features, as well as a flexible framework for Kla prediction. The source code and the training data of the ABFF-Kla and the EBFF-Kla are publicly deposited at: https://github.com/ispotato/Lactylation_model.

OrangeStereo: A navel orange stereo matching network for 3D surface reconstruction

Three-dimensional (3D) surface reconstruction is crucial to accurately detecting navel orange external qualities such as size and shape, and binocular stereo vision is promising for practical sorting scenarios due to its high frame rates and ease of deployment. For objects with texture-less surface, such as navel oranges, it is very challenging to match the corresponding points using traditional stereo matching algorithms because the detected feature points have high similarity in color as well as structure. Most of the existing deep learning-based stereo matching algorithms focus on the stereo matching performance of the whole image and lack the attention to the matching performance of the main region. Meanwhile, simple background with less texture will bring more uncertainty to the matching of corresponding points, which affects the robustness of the model. The OrangeStereo was proposed based on the GwcNet structure to achieve a better corresponding points matching effect of navel oranges, which can provide more accurate 3D surface information. The Structural Features Extraction (SFE) module was introduced after the initial feature extraction module to extract stable structural features and avoid the limitations of using single convolutional features. The Attention Weights Generation (AWG) module was utilized at matching cost calculation stage to help suppress redundant information and make the network pay more attention to the matching performance of specific regions. The improved loss function assigned a greater weight to the loss of the navel orange region using semantic information from left image to make the network more focused on the matching effect of the navel orange region. An ablation study was conducted to confirm the efficiency of the SFE module, AWG module, and the improved loss function of the algorithm. The ablation study contained five sub-experiments, including the baseline model, adding the SFE and the AWG modules respectively, and whether to add the improved loss function based on the addition of previous two modules. Additionally, PSMNet, GwcNet, RealtimeStereo, and ACVNet, four typical stereo matching methods, were contrasted with the OrangeStereo. The proposed OrangeStereo achieved better performance with EPE, RMSE, bad-1, bad-3, bad-5, and inference time of 0.60 pixels, 3.53 pixels, 2.37 %, 1.83 %, 1.62 %, and 33 ms in the orange region, respectively. The R2 and RMSE for the depth of orange utilizing our method were 0.982 and 0.81 mm, respectively. The proposed algorithm can be employed to obtain accurate depth information of fruits quickly, and it is expected to be used in the 3D reconstruction of fruits in the actual commercial fruit sorting lines to achieve more accurate external quality assessment.