Atrous Spatial Pyramid Pooling Research Articles

An Automated Multi-scale Feature Fusion Network for Spine Fracture Segmentation Using Computed Tomography Images.

Spine fractures represent a critical health concern with far-reaching implications for patient care and clinical decision-making. Accurate segmentation of spine fractures from medical images is a crucial task due to its location, shape, type, and severity. Addressing these challenges often requires the use of advanced machine learning and deep learning techniques. In this research, a novel multi-scale feature fusion deep learning model is proposed for the automated spine fracture segmentation using Computed Tomography (CT) to these challenges. The proposed model consists of six modules; Feature Fusion Module (FFM), Squeeze and Excitation (SEM), Atrous Spatial Pyramid Pooling (ASPP), Residual Convolution Block Attention Module (RCBAM), Residual Border Refinement Attention Block (RBRAB), and Local Position Residual Attention Block (LPRAB). These modules are used to apply multi-scale feature fusion, spatial feature extraction, channel-wise feature improvement, segmentation border results border refinement, and positional focus on the region of interest. After that, a decoder network is used to predict the fractured spine. The experimental results show that the proposed approach achieves better accuracy results in solving the above challenges and also performs well compared to the existing segmentation methods.

DSASPP: Depthwise Separable Atrous Spatial Pyramid Pooling for PCB Surface Defect Detection

Printed circuit board (PCB) defect detection is an important and indispensable part of industrial production. PCB defects, due to the small target and similarity between classes, in the actual production of the detection process are prone to omission and false detection problems. Traditional machine-learning-based detection methods are limited by the actual needs of industrial defect detection and do not show good results. Aiming at the problems related to PCB defect detection, we propose a PCB defect detection algorithm based on DSASPP-YOLOv5 and conduct related experiments on the PKU-Market-PCB dataset. DSASPP-YOLOv5 is an improved single-stage detection model, and we first used the K-means++ algorithm for the PKU-Market-PCB dataset to recluster the model so that the model is more in line with the characteristics of PCB small target defects. Second, we design the Depthwise Separable Atrous Spatial Pyramid Pooling (DSASPP) module, which effectively improves the correlation between local and global information by constructing atrous convolution branches with different dilated rates and a global average pooling branch. The experimental results show that our model achieves satisfactory results in both the mean average precision and detection speed metrics compared to existing models, proving the effectiveness of the proposed method.

Segnet with Unet3+ and EfficientNet: a novel framework of brain tumour segmentation and classification model by multiscale attention-based deep learning techniques with hybrid heuristic improvement using 3D MRI brain images

ABSTRACT An adaptive deep learning is recommended to segment and classify the brain tumor using 3D MRI images. Initially, the original 3D MRI images are gathered and fed into pre-processing, which is accomplished using the 'Contrast Limited Adaptive Histogram Equalization (CLAHE)' technique. The pre-processed image is given as input to 'Multiscale Self-guided Attention based Hybrid Adaptive Networks (MSA-HAN)', where it encompasses Segnet with Unet3+ for segmentation. In this model, certain parameters are optimized using the Position Updating of Black Widow and Shark Smell (PUBWSS). Finally, the segmented region is subjected into the Adaptive Atrous Spatial Pyramid Pooling (ASPP) based EfficientNet (AA-ENet) that considers MobileNet and DenseNet. Thus, the model's performance is evaluated, and results are carried out with diverse measures. The performance of the offered method shows 95.74%, and 95.80% in terms of accuracy and F1-score. Hence, the results declare that the developed model achieves higher segmentation accuracy than other approaches.

Deep learning-based cervical lesion segmentation in colposcopic images

Artificial intelligence assisted cancer detection has changed the ream of diagnosis precision. This study aims to propose a segmentation network using artificial intelligence for accurately segmenting the cervix region and acetowhite lesions in cervigram images, addressing the shortage of skilled colposcopists and streamlining the training process. A computational approach is employed to develop and train a deep learning model specifically tailored for cervix region and acetowhite lesion segmentation in cervigram images. A dataset acquired in collaboration with KIDWAI memorial cancer research institute is used for building the model. Cervigram images are collected for training and validation, and a deep learning architecture is constructed and trained using annotated datasets. The segmentation network  based on efficientnet architecture and atrous spatial pyramid pooling is designed to accurately identify and delineate the target regions, with performance evaluation conducted using precision, accuracy, recall, dice score, and specificity metrics. The proposed segmentation network achieves a precision of 0.7387±0.1541, accuracy of 0.9291, recall of 0.7912±0.1439, dice score of 0.7431±0.1506, and specificity of 0.9589±0.0131, indicating its reliability and robustness in segmenting cervix regions and acetowhite lesions in cervigram images. This research demonstrates the feasibility and effectiveness of using artificial intelligence-based computational models for cervix region and acetowhite lesion segmentation in cervigram images. It provides a foundation for further investigations into classifying cervix malignancy using AI techniques, potentially enhancing early detection and treatment of cervical cancer while addressing the shortage of skilled professionals in the field 

A Lightweight 6D Pose Estimation Network Based on Improved Atrous Spatial Pyramid Pooling

It is difficult for lightweight neural networks to produce accurate 6DoF pose estimation effects due to their accuracy being affected by scale changes. To solve this problem, we propose a method with good performance and robustness based on previous research. The enhanced PVNet-based method uses depth-wise convolution to build a lightweight network. In addition, coordinate attention and atrous spatial pyramid pooling are used to ensure accuracy and robustness. This method effectively reduces the network size and computational complexity and is a lightweight 6DoF pose estimation method based on monocular RGB images. Experiments on public datasets and self-built datasets show that the average ADD(-S) estimation accuracy and 2D projection index of the improved method are improved. For datasets with large changes in object scale, the estimation accuracy of the average ADD(-S) is greatly improved.

Prediction of Sea Surface Temperature Using U-Net Based Model

Sea surface temperature (SST) is a key parameter in ocean hydrology. Currently, existing SST prediction methods fail to fully utilize the potential spatial correlation between variables. To address this challenge, we propose a spatiotenporal UNet (ST-UNet) model based on the UNet model. In particular, in the encoding phase of ST-UNet, we use parallel convolution with different kernel sizes to efficiently extract spatial features, and use ConvLSTM to capture temporal features based on the utilization of spatial features. Atrous Spatial Pyramid Pooling (ASPP) module is placed at the bottleneck of the network to further incorporate the multi-scale features, allowing the spatial features to be fully utilized. The final prediction is then generated in the decoding stage using parallel convolution with different kernel sizes similar to the encoding stage. We conducted a series of experiments on the Bohai Sea and Yellow Sea SST data set, as well as the South China Sea SST data set, using SST data from the past 35 days to predict SST data for 1, 3, and 7 days in the future. The model was trained using data spanning from 2010 to 2021, with data from 2022 being utilized to assess the model’s predictive performance. The experimental results show that the model proposed in this research paper achieves excellent results at different prediction scales in both sea areas, and the model consistently outperforms other methods. Specifically, in the Bohai Sea and Yellow Sea sea areas, when the prediction scales are 1, 3, and 7 days, the MAE of ST-UNet outperforms the best results of the other three compared models by 17%, 12%, and 2%, and the MSE by 16%, 18%, and 9%, respectively. In the South China Sea, when the prediction ranges are 1, 3, and 7 days, the MAE of ST-UNet is 27%, 18%, and 3% higher than the best of the other three compared models, and the MSE is 46%, 39%, and 16% higher, respectively. Our results highlight the effectiveness of the ST-UNet model in capturing spatial correlations and accurately predicting SST. The proposed model is expected to improve marine hydrographic studies.

Automatic Liver Segmentation from Multiphase CT Using Modified SegNet and ASPP Module

Liver cancer is one of the dominant causes of cancer death worldwide. Computed Tomography (CT) is the commonly used imaging modality for diagnosing it. Computer-based liver cancer diagnosis systems can assist radiologists in image interpretation and improve diagnosis speed and accuracy. Since liver segmentation is crucial to such systems, researchers are relentlessly pursuing various segmentation approaches. A clinically viable computer-aided system requires examining multiphase CT images. However, most of the research focuses only on the portal venous phase. In this work, we developed an automatic and efficient Deep Learning (DL) method using SegNet, atrous spatial pyramid pooling module and leaky ReLU layers for liver segmentation from quadriphasic abdominal CT volumes. The proposed method was validated on two datasets, an internal institutional dataset consisting of multiphase CT and a public dataset of portal venous phase CT volumes. The Dice Coefficients (DC) obtained were greater than 96% for the latter dataset and the portal venous phase of the former. For arterial, delayed and plain CT phases of the former dataset, the DC achieved were 94.61%, 95.01% and 93.23%, respectively. Experiments showed that our model performed better than the other state-of-the-art DL models. Ablation studies have revealed that the proposed model leverages the strengths of all the three components that make it up. The promising performance of the proposed method suggests that it is appropriate for incorporation in hepatic cancer diagnosis systems.

A hybrid attention-guided ConvNeXt-GRU network for action recognition

In the digital age, with the continuous emergence of large-scale video data, video understanding has become increasingly important. As a core domain, action recognition has garnered widespread attention. However, video exhibits high-dimensional properties and contains human action information at multiple scales, which makes conventional attention mechanisms difficult to capture complex action information. To improve the performance of action recognition, a Hybrid Attention-guided ConvNeXt-GRU Network (HACG) is proposed. Specifically, a Novel Attention Mechanism (ANM) is constructed by integrating a parameter-free attention module into ConvNeXt, enabling the preliminary extraction of important features without the addition of extra parameters. Then, a Multiscale Hybrid Attention Module (MHAM) adopts an improved and efficient Selective Kernel Network (SKNet) to adaptively calibrate channel features. In this way, the module enhances the model’s ability to perceive features at different scales while improving the correlation between channels. Furthermore, MHAM incorporates an Atrous Spatial Pyramid Pooling (ASPP) to extract local and global information from different regions. Finally, MHAM is integrated with the Gated Recurrent Unit (GRU) to capture the interdependence between space and time. Experimental results show that HACG exhibits superior competitiveness compared with the state-of-the-art on the UCF-101, HMDB-51, and Kinetics-400 datasets. This indicates that HACG can more effectively capture important features to suppress noise interference while also having a lower computational load, which makes HACG a highly promising choice for action recognition tasks.

SAB-YOLOv5: An Improved YOLOv5 Model for Permanent Magnetic Ferrite Magnet Rotor Detection

Surface defects on the permanent magnetic ferrite magnet rotor are the primary cause for the decline in performance and safety hazards in permanent magnet motors. Machine-vision methods offer the possibility to identify defects automatically. In response to the challenges in the permanent magnetic ferrite magnet rotor, this study proposes an improved You Only Look Once (YOLO) algorithm named SAB-YOLOv5. Utilizing a line-scan camera, images capturing the complete surface of a general object are obtained, and a dataset containing surface defects is constructed. Simultaneously, an improved YOLOv5-based surface defect algorithm is introduced. Firstly, the algorithm enhances the capability to extract features at different scales by incorporating the Atrous Spatial Pyramid Pooling (ASPP) structure. Then, the fusion of features is improved by combining the tensor concatenation operation of the feature-melting network with the Bidirectional Feature Pyramid Network (BiFPN) structure. Finally, the introduction of the spatial pyramid dilated (SPD) convolutional structure into the backbone network and output end enhances the detection performance for minute defects on the target surface. In the study, the SAB-YOlOv5 algorithm shows an obvious increase from 84.2% to 98.3% in the mean average precision (mAP) compared to that of the original YOLOv5 algorithm. The results demonstrate that the data acquisition method and detection algorithm designed in this paper effectively enhance the efficiency of defect detection permanent magnetic ferrite magnet rotors.

PlaqueNet: deep learning enabled coronary artery plaque segmentation from coronary computed tomography angiography

Cardiovascular disease, primarily caused by atherosclerotic plaque formation, is a significant health concern. The early detection of these plaques is crucial for targeted therapies and reducing the risk of cardiovascular diseases. This study presents PlaqueNet, a solution for segmenting coronary artery plaques from coronary computed tomography angiography (CCTA) images. For feature extraction, the advanced residual net module was utilized, which integrates a deepwise residual optimization module into network branches, enhances feature extraction capabilities, avoiding information loss, and addresses gradient issues during training. To improve segmentation accuracy, a depthwise atrous spatial pyramid pooling based on bicubic efficient channel attention (DASPP-BICECA) module is introduced. The BICECA component amplifies the local feature sensitivity, whereas the DASPP component expands the network’s information-gathering scope, resulting in elevated segmentation accuracy. Additionally, BINet, a module for joint network loss evaluation, is proposed. It optimizes the segmentation model without affecting the segmentation results. When combined with the DASPP-BICECA module, BINet enhances overall efficiency. The CCTA segmentation algorithm proposed in this study outperformed the other three comparative algorithms, achieving an intersection over Union of 87.37%, Dice of 93.26%, accuracy of 93.12%, mean intersection over Union of 93.68%, mean Dice of 96.63%, and mean pixel accuracy value of 96.55%.

Detection Method for Rice Seedling Planting Conditions Based on Image Processing and an Improved YOLOv8n Model

In response to the need for precision and intelligence in the assessment of transplanting machine operation quality, this study addresses challenges such as low accuracy and efficiency associated with manual observation and random field sampling for the evaluation of rice seedling planting conditions. Therefore, in order to build a seedling insertion condition detection system, this study proposes an approach based on the combination of image processing and deep learning. The image processing stage is primarily applied to seedling absence detection, utilizing the centroid detection method to obtain precise coordinates of missing seedlings with an accuracy of 93.7%. In the target recognition stage, an improved YOLOv8 Nano network model is introduced, leveraging deep learning algorithms to detect qualified and misplaced seedlings. This model incorporates ASPP (atrous spatial pyramid pooling) to enhance the network’s multiscale feature extraction capabilities, integrates SimAM (Simple, Parameter-free Attention Module) to improve the model’s ability to extract detailed seedling features, and introduces AFPN (Asymptotic Feature Pyramid Network) to facilitate direct interaction between non-adjacent hierarchical levels, thereby enhancing feature fusion efficiency. Experimental results demonstrate that the enhanced YOLOv8n model achieves precision (P), recall (R), and mean average precision (mAP) of 95.5%, 92.7%, and 95.2%, respectively. Compared to the original YOLOv8n model, the enhanced model shows improvements of 3.6%, 0.9%, and 1.7% in P, R, and mAP, respectively. This research provides data support for the efficiency and quality of transplanting machine operations, contributing to the further development and application of unmanned field management in subsequent rice seedling cultivation.

Multi-view orientational attention network combining point-based affinity for polyp segmentation

Most existing deep learning-based polyp segmentation methods neglect two important aspects of polyps: the geometric orientation information of polyps and the point information of the entire colonoscopy area. In this paper, we introduce a multi-view orientational attention network (MVOA-Net), which incorporates orientation and point awareness to effectively address the issue of intra-class inconsistency resulting from variations in polyp shape, size, and position, as well as the inter-class indistinction caused by the high similarity between polyp lesions and surrounding tissues. To achieve robust orientation awareness, we propose a novel geometric orientation transformer encoder (GOTE) based on horizontal and vertical views. Moreover, To simultaneously capture the global context information of GOTE and emphasize the important local information of the convolution-based attention encoder (CBAE), a global and local cross attention fusion module (CAFM) is also proposed to simultaneously model the long-range dependencies of polyps and pay sufficient attention to the local boundaries of polyps. Additionally, a efficient atrous spatial pyramid pooling (E-ASPP) module is proposed to enhance the semantic representation of high-level features. Finally, a point-based affinity module (PBAM) and a multi-scale fusion module (MSFM) are proposed to distinguish the disguise of polyps, further alleviating inter-class indistinction. The ablation study results demonstrate the effectiveness of each component. Quantitative and qualitative experimental results show that MVOA-Net achieves the best segmentation accuracy across domain polyp datasets and has obvious advantages in segmenting multiple polyp objects.

Robust residual convolutional neural network based pupil tracking for low-computational power applications

Existing Deep Fully Convolutional Network (DFCN) based pupil segmentation models have been shown to perform robust and accurate pupil detection. However, they lack high computational resources when real-time predictions for low computational wearable devices such as Raspberry Pi are of interest. Simple FCN (Fully Convolutional Network) models, on the other hand, can provide real-time yet not as accurate predictions with low computational power devices. In this study, we address this dilemma by proposing a CNN (Convolutional Neural Network)-based model, Residual CNN, with several advanced operations such as residual connections, Squeeze and Excitation (SE) attention, and Atrous Spatial Pyramid Pooling (ASPP) to improve the prediction performance without complicating the model. Moreover, we leverage transfer learning by training synthetic images, then fine-tuning them with authentic eye images. We also fully quantize our model parameters to speed up the predictions and apply the Quantization Aware Training (QAT) strategy to obtain accurate predictions. Our experiments show that our quantized Res-CNN model trained by QAT strategy with 40x30 resolution images provides robust yet real-time predictions with an average of 1.351 Root Mean Square Error (RMSE) for the pupil center predictions and 8.317 ms response time per image on Raspberry Pi.

RAAWC-UNet: an apple leaf and disease segmentation method based on residual attention and atrous spatial pyramid pooling improved UNet with weight compression loss.

Early detection of leaf diseases is necessary to control the spread of plant diseases, and one of the important steps is the segmentation of leaf and disease images. The uneven light and leaf overlap in complex situations make segmentation of leaves and diseases quite difficult. Moreover, the significant differences in ratios of leaf and disease pixels results in a challenge in identifying diseases. To solve the above issues, the residual attention mechanism combined with atrous spatial pyramid pooling and weight compression loss of UNet is proposed, which is named RAAWC-UNet. Firstly, weights compression loss is a method that introduces a modulation factor in front of the cross-entropy loss, aiming at solving the problem of the imbalance between foreground and background pixels. Secondly, the residual network and the convolutional block attention module are combined to form Res_CBAM. It can accurately localize pixels at the edge of the disease and alleviate the vanishing of gradient and semantic information from downsampling. Finally, in the last layer of downsampling, the atrous spatial pyramid pooling is used instead of two convolutions to solve the problem of insufficient spatial context information. The experimental results show that the proposed RAAWC-UNet increases the intersection over union in leaf and disease segmentation by 1.91% and 5.61%, and the pixel accuracy of disease by 4.65% compared with UNet. The effectiveness of the proposed method was further verified by the better results in comparison with deep learning methods with similar network architectures.

Multi-Scale Feature Fusion Network with Symmetric Attention for Land Cover Classification Using SAR and Optical Images

The complementary characteristics of SAR and optical images are beneficial in improving the accuracy of land cover classification. Deep learning-based models have achieved some notable results. However, how to effectively extract and fuse the unique features of multi-modal images for pixel-level classification remains challenging. In this article, a two-branch supervised semantic segmentation framework without any pretrained backbone is proposed. Specifically, a novel symmetric attention module is designed with improved strip pooling. The multiple long receptive fields can better perceive irregular objects and obtain more anisotropic contextual information. Meanwhile, to solve the semantic absence and inconsistency of different modalities, we construct a multi-scale fusion module, which is composed of atrous spatial pyramid pooling, varisized convolutions and skip connections. A joint loss function is introduced to constrain the backpropagation and reduce the impact of class imbalance. Validation experiments were implemented on the DFC2020 and WHU-OPT-SAR datasets. The proposed model achieved the best quantitative values on the metrics of OA, Kappa and mIoU, and its class accuracy was also excellent. It is worth mentioning that the number of parameters and the computational complexity of the method are relatively low. The adaptability of the model was verified on RGB–thermal segmentation task.

Depth_ASPP_UNet++: a novel 2D ESPI phase unwrapping based on convolutional neural network

Phase unwrapping is a key step in optical metrology and physical optics to obtain accurate phase distributions. In practice, phase images obtained from electronic speckle pattern interferometry (ESPI) exhibit diverse and complex morphology, with significant shape variations and non-uniform densities among different individuals. This takes challenges for accurately extracting phase information and unwrapping the phase. With the progress of deep learning technology in optical image processing, real-time performance and accuracy have become concerned issues. In this paper, an ESPI phase unwrapping method based on convolutional neural network UNet++ is proposed. The proposed network combines the depthwise separable convolution (DSC), atrous spatial pyramid pooling (ASPP), defined as Depth_ASPP_UNet++. In this model, the use of DSC improves network computational efficiency and provides better feature representation capability. In addition, ASPP is introduced to pay more attention to the phase information of the phase image, and then obtain better phase unwrapping results. The experimental results show that our proposed method can obtain excellent results, especially with various of variable density, different noise levels, and different speckle sizes.

Usage of an improved YOLOv5 for steel surface defect detection

Abstract The one-stage YOLOv5 steel surface defect detection has issues such as slow operation speed, loss of defect location and semantic information of small targets, and inadequate extraction of defect features. This study proposed a defect detection algorithm with improved YOLOv5 to solve these issues. The proposed algorithm used the slim-neck layer built by three new modules instead of the neck layer in YOLOv5s to achieve a lightweight network model. In addition, the spatial perception self-attention mechanism was introduced to enhance the feature extraction capability of the initial convolutional layer without limiting the input size. The improved Atrous Spatial Pyramid Pooling was added to expand the perceptual field and capture multiscale contextual information while preventing local information loss and enhancing the relevance of long-range information. The experimental results showed that the improved YOLOv5 algorithm has a reduced model volume, significantly higher detection accuracy and speed than the traditional algorithm, and the ability to detect steel surface defects quickly and accurately.

Multi-class segmentation of navel orange surface defects based on improved DeepLabv3+

To address the problems of current mainstream semantic segmentation network such as rough edge segmentation of navel oranges defects, poor accuracy of small target defect segmentation and insufficient deep-level semantic extraction of defects, feature information will be lost, a multi-class segmentation model based on improved DeepLabv3+ is proposed to detect the surface defects of navel oranges. The Coordinate Attention Mechanism is embedded into the DeepLabv3+ network for better semantic segmentation performance, while the dilated convolution of Atrous Spatial Pyramid Pooling structure is replaced with deformable empty convolution to improve the fitting ability of the network to target shape changes and irregular defects. In addition, a BiFPN-based feature fusion branch is introduced at the DeepLabv3+ encoder side to realize multi-scale feature fusion and enrich feature space and semantic information. The experimental results show that the average intersection ratio and average pixel intersection ratio accuracies of the improved DeepLabv3+ model on the navel orange surface defect dataset are 77.32% and 86.38%, which are 3.81% and 5.29% higher than the original DeepLabv3+ network, respectively, improving the extraction capability of navel orange defect features and having better segmentation performance.

Urban Vegetation Classification for Unmanned Aerial Vehicle Remote Sensing Combining Feature Engineering and Improved DeepLabV3+

Addressing the problems of misclassification and omissions in urban vegetation fine classification from current remote sensing classification methods, this research proposes an intelligent urban vegetation classification method that combines feature engineering and improved DeepLabV3+ based on unmanned aerial vehicle visible spectrum images. The method constructs feature engineering under the ReliefF algorithm to increase the number of features in the samples, enabling the deep learning model to learn more detailed information about the vegetation. Moreover, the method improves the classical DeepLabV3+ network structure based on (1) replacing the backbone network using MoblieNetV2; (2) adjusting the atrous spatial pyramid pooling null rate; and (3) adding the attention mechanism and the convolutional block attention module. Experiments were conducted with self-constructed sample datasets, where the method was compared and analyzed with a fully convolutional network (FCN) and U-Net and ShuffleNetV2 networks; the migration of the method was tested as well. The results show that the method in this paper is better than FCN, U-Net, and ShuffleNetV2, and reaches 92.27%, 91.48%, and 85.63% on the accuracy evaluation indices of overall accuracy, MarcoF1, and mean intersection over union, respectively. Furthermore, the segmentation results are accurate and complete, which effectively alleviates misclassifications and omissions of urban vegetation; moreover, it has a certain migration ability that can quickly and accurately classify the vegetation.

Ancient mural segmentation based on multiscale feature fusion and dual attention enhancement

To address the fuzzy segmentation boundaries, missing details, small target losses and low efficiency of traditional segmentation methods in ancient mural image segmentation scenarios, this paper proposes a mural segmentation model based on multiscale feature fusion and a dual attention-augmented segmentation model (MFAM). The model uses the MobileViT network, which integrates a coordinate attention mechanism, as the feature extraction backbone network. It attains global and local expression capabilities through self-attention, class convolution, and coordinate attention and focuses on location information to expand the receptive field and achieve improved feature extraction efficiency. An A_R_ASPP feature enhancement module is proposed for the attention-optimized residual atrous spatial pyramid pooling module. The module uses residual connections to solve the small target loss problem in murals caused by the excessive sampling rate of atrous convolution and uses a feature attention mechanism to adaptively adjust the feature map weight according to the channel importance levels. A dual attention-enhanced feature fusion module is proposed for multiscale decoder feature fusion to improve the mural segmentation effect. This module uses a cross-level aggregation strategy and an attention mechanism to weight the importance of different feature levels to obtain multilevel semantic feature representations. The model improves the mean intersection over union (MIoU) by 3.06% and the MPA by 1.81% on a mural dataset compared with other models. The model is proven to be effective at improving the segmentation details, efficiency and small target segmentation results produced for mural images, and a new method is proposed for segmenting ancient mural images.