Residual Modulation Research Articles

Lightweight monocular depth estimation using a fusion-improved transformer

The existing deep estimation networks often overlook the issue of computational efficiency while pursuing high accuracy. This paper proposes a lightweight self-supervised network that combines convolutional neural networks (CNN) and Transformers as the feature extraction and encoding layers for images, enabling the network to capture both local geometric and global semantic features for depth estimation. First, depth-separable convolution is used to construct a dilated convolution residual module based on a shallow network to improve the shallow CNN feature extraction receptive field. In the transformer, a multidepth separable convolution head transposed attention module is proposed to reduce the computational burden of spatial self-attention. In the feedforward network, a two-step gating mechanism is proposed to improve the nonlinear representation ability of the feedforward network. Finally, the CNN and transformer are integrated to implement a depth estimation network with a local-global context interaction function. Compared with other lightweight models, this model has fewer model parameters and higher estimation accuracy. It also has better generalizability for different outdoor datasets. Additionally, the inference speed can reach 87 FPS, achieving better real-time performance and accounting for both inference speed and estimation accuracy.

ResMT: A hybrid CNN-transformer framework for glioma grading with 3D MRI

Accurate grading of gliomas is crucial for treatment strategies and prognosis. While convolutional neural networks (CNNs) have proven effective in classifying medical images, they struggle with capturing long-range dependencies among pixels. Transformer-based networks can address this issue, but CNN-based methods often perform better when trained on small datasets. Additionally, tumor segmentation is essential for classification models, but training an additional segmentation model significantly increases workload. To address these challenges, we propose ResMT, which combines CNN and transformer architectures for glioma grading, extracting both local and global features efficiently. Specifically, we designed a spatial residual module (SRM) where a 3D CNN captures glioma's volumetric complexity, and Swin UNETR, a pre-trained segmentation model, enhances the network without extra training. Our model also includes a multi-plane channel and spatial attention module (MCSA) to refine the analysis by focusing on critical features across multiple planes (axial, coronal, and sagittal). Transformer blocks establish long-range relationships among planes and slices. We evaluated ResMT on the BraTs19 dataset, comparing it with baselines and state-of-the-art models. Results demonstrate that ResMT achieves the highest prediction performance with an AUC of 0.9953, highlighting hybrid CNN-transformer models' potential for 3D MRI classification.

RAM-YOLOv8: A multi-fault detection method for photovoltaic module

ABSTRACT Given the constraints in fault type identification, localization precision, and detection efficiency inherent in current photovoltaic fault detection methodologies, this paper introduces an innovative detection approach integrating RAM-YOLOv8 with the Swin Transformer. To mitigate the limitations in model extraction capabilities during the detection process and circumvent the vanishing gradient issue encountered during backward information propagation, this study introduces a residual aggregation module as a substitute for the original C2F module, thereby augmenting feature extraction efficiency. Furthermore, within the feature fusion network, the Swin attention mechanism fusion module is designed by integrating the Swin Transformer’s architectural concept to minimize the loss of weak defect information and suppress irrelevant background interference. The model achieves a detection accuracy of 88.5% on the dataset, representing a 7% improvement over the original YOLOv8 algorithm, with a detection rate of 83.33 FPS on an RTX2080 GPU. This method offers a robust and efficient fault detection solution for PV systems, rendering it a practical approach for enhancing system performance and reducing maintenance costs.

Dunhuang mural inpainting based on reference guidance and multi‐scale fusion

AbstractIn response to the inadequate utilization of prior information in current mural inpainting processes, leading to issues such as semantically unreliable inpaintings and the presence of artifacts in the inpainting area, a Dunhuang mural inpainting method based on reference guidance and multi‐scale feature fusion is proposed. First, the simulated broken mural, the mask image, and the reference mural are input into the model to complete the multi‐level embedding of patches and align the multi‐scale fine‐grained features of damaged murals and reference murals. Following the patch embedding module, a hybrid residual module is added based on hybrid attention to fully extract mural features. In addition, by continuing the residual concatenation of outputs of the hierarchical embedding module improves the ability of the model to represent deeper features, and improves the robustness and generalisation of the model. Second, the encoded features are fed into the decoder to generate decoded features. Finally, the convolutional tail is employed to propagate them and complete the mural painting. Experimental validation on the Dunhuang mural dataset demonstrates that, compared to other algorithms, this model exhibits higher evaluation metrics in the inpainting of extensively damaged murals and demonstrates overall robustness. In terms of visual effects, the results of this model in the inpainting process exhibit finer textures, richer semantic information, more coherent edge structures, and a closer resemblance to authentic murals.

DSC-Net: Enhancing Blind Road Semantic Segmentation with Visual Sensor Using a Dual-Branch Swin-CNN Architecture.

In modern urban environments, visual sensors are crucial for enhancing the functionality of navigation systems, particularly for devices designed for visually impaired individuals. The high-resolution images captured by these sensors form the basis for understanding the surrounding environment and identifying key landmarks. However, the core challenge in the semantic segmentation of blind roads lies in the effective extraction of global context and edge features. Most existing methods rely on Convolutional Neural Networks (CNNs), whose inherent inductive biases limit their ability to capture global context and accurately detect discontinuous features such as gaps and obstructions in blind roads. To overcome these limitations, we introduce Dual-Branch Swin-CNN Net(DSC-Net), a new method that integrates the global modeling capabilities of the Swin-Transformer with the CNN-based U-Net architecture. This combination allows for the hierarchical extraction of both fine and coarse features. First, the Spatial Blending Module (SBM) mitigates blurring of target information caused by object occlusion to enhance accuracy. The hybrid attention module (HAM), embedded within the Inverted Residual Module (IRM), sharpens the detection of blind road boundaries, while the IRM improves the speed of network processing. In tests on a specialized dataset designed for blind road semantic segmentation in real-world scenarios, our method achieved an impressive mIoU of 97.72%. Additionally, it demonstrated exceptional performance on other public datasets.

A fully locally selective large kernel network for traffic video detection

Vision Transformer (ViT) are widely used in the field of traffic video inspection because of their wide sensing field and strong global context-capturing capability. However, the ViT model has inadequate local detail feature extraction and high computational complexity. Convoluted neural networks(CNN) are commonly used as a means of video image detection. Although it is more flexible and less computationally complex in the extraction of local details, the fixed-size kernel limits it. It is difficult to effectively utilize the full local context information, resulting in the difficulty of effectively extracting the tiny targets with uneven scale distribution in the region. It is prone to leakage and misdetection of small targets. In view of this, this study proposes a new fully localized selective large kernel network (FL-SLKNet). The network effectively captures specific targets and background information through a fully localized feature extraction method, thus making small targets in the region more distinctive features. On this basis, the Adaptive Expansion Residual Module (AERM) is proposed, which solves the problem of image resolution degradation due to sensory field expansion by combining point-by-point convolution and expansion convolution and utilizing residual connections to supplement the lost information. In order to effectively identify tiny targets in the local area, this study proposes the Multi-Scale Frequency Domain Encoder (MSFDEncoder); this encoder utilizes a high-frequency branch to capture local details and a low-frequency branch to focus on the global structure, effectively capturing granular features of targets at various scales. Experimental results on the VisDrone2019-DET and BDD-100K datasets indicate that FL-SLKNet outperforms other advanced models in traffic video detection, with an increase of 3.3% and 3.4% in mAP0.5 compared to the RT-DETR model. Meanwhile, the detection performance of FL-SLKNet is excellent compared with the YOLO series on the SZ Actual Traffic Video dataset. In addition, in the rainy and hazy simulation scenarios, FL-SLKNet’s mAP0.5 improves by 5.5% and 2.6% compared to the RT-DETR model, and the extreme weather robustness performance is better.

TAFENet: A Two-Stage Attention-Based Feature-Enhancement Network for Strip Steel Surface Defect Detection

Strip steel serves as a crucial raw material in numerous industries, including aircraft and automobile manufacturing. Surface defects in strip steel can degrade the performance, quality, and appearance of industrial steel products. Detecting surface defects in steel strip products is challenging due to the low contrast between defects and background, small defect targets, as well as significant variations in defect sizes. To address these challenges, a two-stage attention-based feature-enhancement network (TAFENet) is proposed, wherein the first-stage feature-enhancement procedure utilizes an attentional convolutional fusion module with convolution to combine all four-level features and then strengthens the features of different levels via a residual spatial-channel attention connection module (RSC). The second-stage feature-enhancement procedure combines three-level features using an attentional self-attention fusion module and then strengthens the features using a RSC attention module. Experiments on the NEU-DET and GC10-DET datasets demonstrated that the proposed method significantly improved detection accuracy, thereby confirming the effectiveness and generalization capability of the proposed method.

3D Object Detection via Residual SqueezeDet

Three-dimensional object detection is a critical task in computer vision with applications in autonomous driving, robotics, and augmented reality. This paper introduces Residual SqueezeDet, a novel network architecture that enhances the performance of 3D object detection on the KITTI dataset. Building upon the efficient SqueezeDet framework, we propose the Residual Fire module, which incorporates skip connections inspired by ResNet architectures into the original Fire module. This innovation improves gradient flow, enhances feature propagation, and allows for more effective training of deeper networks. Our method leverages point cloud and image-based features, employing a Residual SqueezeDet to effectively capture local and global context. Extensive experiments on the KITTI dataset demonstrate that Residual SqueezeDet significantly outperforms the original SqueezeDet, with particularly notable improvements in challenging scenarios. The proposed model maintains computational efficiency while achieving state-of-the-art performance, making it well-suited for real-time applications in autonomous driving. Our work contributes to the field by providing a more accurate and robust solution for 3D object detection, paving the way for improved perception systems in dynamic environments.

CP-RDM: a new object detection algorithm for casting and pouring robots

Abstract Automating the casting sector heavily relies on pivotal technology for object detection in pouring robots. A sophisticated algorithm designed to identify and locate target pouring holes in intricate casting workshops is crucial for advancing the intelligence of the casting process. However, the workshop environment for pouring is generally challenging, with uneven lighting, varying sizes of pouring holes, and significant occlusion in the target area, all impacting the accuracy of target detection tasks.
To overcome these challenges, this paper proposes enhancing the YOLOv8s algorithm for object detection in pouring robots. Firstly, to address the issue of different scales in pouring holes, a Multi-Scale Residual Channel and Spatial Information Fusion Module (MRCS) is designed to aggregate channel and spatial information, thereby enhancing the feature extraction capability of the model. The proposed enhancement is validated on the Pascal VOC dataset. Secondly, a SimAM attention mechanism is added at the end of the backbone network to focus the object detection network more on the positional region of the pouring hole. Importantly, this addition does not introduce extra parameters or computational burden to the model.
Finally, in the detection part of the model, the detection head from the RT-DETR model is introduced. This combination of real-time detection capability from YOLO and deep feature extraction capability from RT-DETR enhances the detection accuracy of the model while ensuring real-time performance. Experimental results on the updated pouring hole dataset reveal that, with only a slight increase in parameters, the proposed model achieves a 2.5% and 3.5% improvement in mAP@0.5 and F1-Score, respectively, compared to the baseline algorithm YOLOv8s. Precision (P) is enhanced by 1.8%, recall (R) by 3.5%, and PFS reaches 110, meeting the requirements for real-time pouring in pouring robots.

Luminance decomposition and reconstruction for high dynamic range Video Quality Assessment

High dynamic range (HDR) video represents a wider range of brightness, detail and colour than standard dynamic range (SDR) video. However, SDR-based VQA (Video Quality Assessment) models struggle to capture HDR distortions. In addition, some of the existing methods designed for HDR video focus on emphasising the distortion of local areas of the video frame, ignoring the distortion of the video frame as a whole. Therefore, we propose a no reference VQA model based on luminance decomposition and recombination that provides excellent performance for HDR videos, called HDR-DRVQA. Specifically, HDR-DRVQA utilises a luminance decomposition strategy to decompose video frames into different regions for explicit extraction of perceptual features in different regions of the high dynamic range. We then further propose a residual aggregation module for recombining multi-region features to extract static spatial distortion representations and dynamic motion perception (captured by feature differences). Taking advantage of the Transformer network in remote dependency modelling, this information is fed into the Transformer network for interactive learning of motion perception and adaptively constructs a stream of spatial distortion information from shallow to deep layers during temporal aggregation. We validate that our model significantly outperforms SDR VQA and existing HDR VQA methods on the publicly available HDR databases.

A Multi-Scale Liver Tumor Segmentation Method Based on Residual and Hybrid Attention Enhanced Network with Contextual Integration.

Liver cancer is one of the malignancies with high mortality rates worldwide, and its timely detection and accurate diagnosis are crucial for improving patient prognosis. To address the limitations of traditional image segmentation techniques and the U-Net network in capturing fine image features, this study proposes an improved model based on the U-Net architecture, named RHEU-Net. By replacing traditional convolution modules in the encoder and decoder with improved residual modules, the network's feature extraction capabilities and gradient stability are enhanced. A Hybrid Gated Attention (HGA) module is integrated before the skip connections, enabling the parallel processing of channel and spatial attentions, optimizing the feature fusion strategy, and effectively replenishing image details. A Multi-Scale Feature Enhancement (MSFE) layer is introduced at the bottleneck, utilizing multi-scale feature extraction technology to further enhance the expression of receptive fields and contextual information, improving the overall feature representation effect. Testing on the LiTS2017 dataset demonstrated that RHEU-Net achieved Dice scores of 95.72% for liver segmentation and 70.19% for tumor segmentation. These results validate the effectiveness of RHEU-Net and underscore its potential for clinical application.

CENN: Capsule-enhanced neural network with innovative metrics for robust speech emotion recognition

Speech emotion recognition (SER) plays a pivotal role in enhancing Human-computer interaction (HCI) systems. This paper introduces a groundbreaking Capsule-enhanced neural network (CENN) that significantly advances the state of SER through a robust and reproducible deep learning framework. The CENN architecture seamlessly integrates advanced components, including Multi-head attention (MHA), residual module, and capsule module, which collectively enhance the model's capacity to capture both global and local features essential for precise emotion classification. A key contribution of this work is the development of a comprehensive reproducibility framework, featuring novel metrics: General learning reproducibility (GLR) and Correct learning reproducibility (CLR). These metrics, alongside their fractional and perfect variants, offer a multi-dimensional evaluation of the model's consistency and correctness across multiple executions, thereby ensuring the reliability and credibility of the results. To tackle the persistent challenge of overfitting in deep learning models, we propose an innovative overfitting metric that considers the intricate relationship between training and testing errors, model complexity, and data complexity. This metric, in conjunction with the newly introduced generalization and robustness metrics, provides a holistic assessment of the model's performance, guiding the application of regularization techniques to maintain generalizability and resilience. Extensive experiments conducted on benchmark SER datasets demonstrate that the CENN model not only surpasses existing approaches in terms of accuracy but also sets a new benchmark in reproducibility. This work establishes a new paradigm for deep learning model development in SER, underscoring the vital importance of reproducibility and offering a rigorous framework for future research.

Brain tumor image segmentation method using hybrid attention module and improved mask RCNN

To meet the needs of automated medical analysis of brain tumor magnetic resonance imaging, this study introduces an enhanced instance segmentation method built upon mask region-based convolutional neural network. By incorporating squeeze-and-excitation networks, a channel attention mechanism, and concatenated attention neural network, a spatial attention mechanism, the model can more adeptly focus on the critical regions and finer details of brain tumors. Residual network-50 combined attention module and feature pyramid network as the backbone network to effectively capture multi-scale characteristics of brain tumors. At the same time, the region proposal network and region of interest align technology were used to ensure that the segmentation area matched the actual tumor morphology. The originality of the research lies in the deep residual network that combines attention mechanism with feature pyramid network to replace the backbone based on mask region convolutional neural network, achieving an improvement in the efficiency of brain tumor feature extraction. After a series of experiments, the precision of the model is 90.72%, which is 0.76% higher than that of the original model. Recall was 91.68%, an increase of 0.95%; Mean Intersection over Union was 94.56%, an increase of 1.39%. This method achieves precise segmentation of brain tumor magnetic resonance imaging, and doctors can easily and accurately locate the tumor area through the segmentation results, thereby quickly measuring the diameter, area, and other information of the tumor, providing doctors with more comprehensive diagnostic information.

An Effective Res-Progressive Growing Generative Adversarial Network-Based Cross-Platform Super-Resolution Reconstruction Method for Drone and Satellite Images

In recent years, accurate field monitoring has been a research hotspot in the domains of aerial remote sensing and satellite remote sensing. In view of this, this study proposes an innovative cross-platform super-resolution reconstruction method for remote sensing images for the first time, aiming to make medium-resolution satellites capable of field-level detection through a super-resolution reconstruction technique. The progressive growing generative adversarial network (PGGAN) model, which has excellent high-resolution generation and style transfer capabilities, is combined with a deep residual network, forming the Res-PGGAN model for cross-platform super-resolution reconstruction. The Res-PGGAN architecture is similar to that of the PGGAN, but includes a deep residual module. The proposed Res-PGGAN model has two main benefits. First, the residual module facilitates the training of deep networks, as well as the extraction of deep features. Second, the PGGAN structure performs well in cross-platform sensor style transfer, allowing for cross-platform high-magnification super-resolution tasks to be performed well. A large pre-training dataset and real data are used to train the Res-PGGAN to improve the resolution of Sentinel-2’s 10 m resolution satellite images to 0.625 m. Three evaluation metrics, including the structural similarity index metric (SSIM), the peak signal-to-noise ratio (PSNR), and the universal quality index (UQI), are used to evaluate the high-magnification images obtained by the proposed method. The images generated by the proposed method are also compared with those obtained by the traditional bicubic method and two deep learning super-resolution reconstruction methods: the enhanced super-resolution generative adversarial network (ESRGAN) and the PGGAN. The results indicate that the proposed method outperforms all the comparison methods and demonstrates an acceptable performance regarding all three metrics (SSIM/PSNR/UQI: 0.9726/44.7971/0.0417), proving the feasibility of cross-platform super-resolution image recovery.

A Self-Supervised Network-Based Smoke Removal and Depth Estimation for Monocular Endoscopic Videos.

In minimally invasive surgery videos, label-free monocular laparoscopic depth estimation is challenging due to smoke. For this reason, we propose a self-supervised collaborative network-based depth estimation method with smoke-removal for monocular endoscopic video, which is decomposed into two steps of smoke-removal and depth estimation. In the first step, we develop a de-endoscopic smoke for cyclic GAN (DS-cGAN) to mitigate the smoke components at different concentrations. The designed generator network comprises sharpened guide encoding module (SGEM), residual dense bottleneck module (RDBM) and refined upsampling convolution module (RUCM), which restores more detailed organ edges and tissue structures. In the second step, high resolution residual U-Net (HRR-UNet) consisting of a DepthNet and two PoseNets is designed to improve the depth estimation accuracy, and adjacent frames are used for camera self-motion estimation. In particular, the proposed method requires neither manual labeling nor patient computed tomography scans during the training and inference phases. Experimental studies on the laparoscopic data set of the Hamlyn Centre show that our method can effectively achieve accurate depth information after net smoking in real surgical scenes while preserving the blood vessels, contours and textures of the surgical site. The experimental results demonstrate that the proposed method outperforms existing state-of-the-art methods in effectiveness and achieves a frame rate of 94.45fps in real time, making it a promising clinical application.

A Multi-Branch Feature Extraction Residual Network for Lightweight Image Super-Resolution

Single-image super-resolution (SISR) seeks to elucidate the mapping relationships between low-resolution and high-resolution images. However, high-performance network models often entail a significant number of parameters and computations, presenting limitations in practical applications. Therefore, prioritizing a light weight and efficiency becomes crucial when applying image super-resolution (SR) to real-world scenarios. We propose a straightforward and efficient method, the Multi-Branch Feature Extraction Residual Network (MFERN), to tackle lightweight image SR through the fusion of multi-information self-calibration and multi-attention information. Specifically, we have devised a Multi-Branch Residual Feature Fusion Module (MRFFM) that leverages a multi-branch residual structure to succinctly and effectively fuse multiple pieces of information. Within the MRFFM, we have designed the Multi-Scale Attention Feature Fusion Block (MAFFB) to adeptly extract features via convolution and self-calibration attention operations. Furthermore, we introduce a Dual Feature Calibration Block (DFCB) to dynamically fuse feature information using dynamic weight values derived from the upper and lower branches. Additionally, to overcome the limitation of convolution in solely extracting local information, we incorporate a Transformer module to effectively integrate global information. The experimental results demonstrate that our MFERN exhibits outstanding performance in terms of model parameters and overall performance.

Apple recognition in complex environments based on FC-DETR

This study aims to address the challenges of apple recognition in complex environments by proposing a solution based on the FC-DETR model. Apple cultivation is an important part of agriculture in Xinjiang, but the increasing shortage of labor has driven the demand for automated harvesting technologies. Therefore, research and development of apple recognition technology in complex environments have become crucial. While CNN architecture models can accurately identify apples in conventional settings, their inherent receptive field limitations prevent them from fully capturing global features in noisy and complex environments, making it difficult to achieve the accuracy and robustness required for practical applications. To solve this problem, this study proposes a real-time object detection model, FC-DETR. This model incorporates the innovative FEMA-BasicBlock residual module, the CAFM cross-scale adaptive feature fusion module, and the novel Inner-WIoU loss function, thereby enhancing feature processing, multi-scale feature selection and integration, and detection accuracy. Experimental results show that the FC-DETR model achieves an apple recognition accuracy of 87 % and a recall rate of 82 % in complex backgrounds while maintaining a lightweight design. This study not only makes significant advances in automated apple harvesting technology but also contributes to improving the efficiency and sustainability of the apple industry in Xinjiang and globally.

MRCFN: A multi-sensor residual convolutional fusion network for intelligent fault diagnosis of bearings in noisy and small sample scenarios

Bearing fault diagnosis is of great importance to ensure the safe and stable operation of mechanical equipment. The actual collected bearing fault signals are susceptible to strong noise interference and bearing samples for each fault state may be insufficient, which increases the difficulty of capturing effective features. Most of the existing diagnostic methods extract features from a single sensor signal for pattern recognition and fault diagnosis. The fault information provided by a single sensor is limited and incomplete, which is usually very difficult to meet the demand for accurate and reliable fault diagnosis in complex scenarios. To solve these problems, this paper proposes a multi-sensor residual convolutional fusion network (MRCFN) for intelligent fault diagnosis of bearings. Firstly, a convolutional pooling module (CPM) is coupled with the designed double ring residual module (DRRM) to rough feature extraction and deep feature mining, which not only captures the discriminative fault features from multi-sensor signal, but also avoids the performance degradation of network. Secondly, a spatial channel reconstruction module (SCRM) is further introduced to eliminate redundant information in the features and improve the network training efficiency. Finally, the presented global interactive perception fusion module (GIPFM) is connected with a classification block (CB) to globally fuse the features extracted from the acoustic and vibration signals and conduct automatic fault identification, which both can realize the complementarity and calibration of multi-sensor feature information and high precision diagnosis. The experiments and a series of comparisons are implemented on two datasets to efficaciously verify the superiority of the proposed method over five existing representative multi-sensor fusion diagnosis methods (i.e., FAC-CNN, MB-CNN, MsACNN, MRSDF, and IMSFDFL) under strong noise and small samples.

FCT-Net: A dual-encoding-path network fusing atrous spatial pyramid pooling and transformer for pavement crack detection

Cracks are a typical form of road damage, and accurate detection of cracks is of great significance for road maintenance work and ensuring traffic safety. Recently, computer vision has gradually been applied in the field of crack segmentation. However, there are still some extremely challenging problems in crack segmentation, such as complex backgrounds, information loss caused by pooling and convolution operations, and insufficient fusion of global and local semantic information. In response to the above problems, this paper proposes a dual-encoding-path network with U-Net architecture called FCT-Net, by fusing channel atrous spatial pyramid pooling (CASPP) and transformer. Specifically, CASPP obtains multi-scale receptive fields by incorporating spatial and channel attention, while refining and extracting local features. Meanwhile, we introduce long-short distance attention to construct a novel transformer with the prominent characteristic of interaction between local and global attention features. In addition, a residual convolution module is designed to enhance the local features of the transformer. Furthermore, we devise a multi-scale attention weight cross fusion module to aggregate the features of the dual encoding branch, for reducing information loss during downsampling and suppress background information. Eventually, we evaluate the performance of FCT-Net by experiments on three public datasets. Extensive experimental results show that FCT-Net achieves higher F1-score and mean intersection over union (mIoU) than state-of-the-art segmentation networks on the DeepCrack537 and CrackLS315 datasets. Meanwhile, it has excellent segmentation performance for cracks in complex scenes, with the highest recall, F1-score, and mIoU respectively as 85.64%, 81.67%, and 84.05% on the CrackTree260 dataset.

Automatic Segmentation Model for Parkinson’s Images Based on SA-U2-Net

To effectively solve the problem of a small proportion of substantia nigra and midbrain regions in Magnetic Resolution Imaging (MRI) images of Parkinson’s disease (PD) patients, unclear boundaries with surrounding tissues, and difficulty in accurately delineating boundaries, an improved U2-Net algorithm for Parkinson’s substantia nigra midbrain image segmentation was proposed. This algorithm first improves the Residual U-blocks (RSU) and RSU-4F modules using the Shuffle Attention (SA) module, enhancing the network’s attention to Parkinson’s substantia nigra and midbrain blurry regions in the encoding and decoding layers. Next, replace some ordinary convolutions in RSU-4F with DynamicConv (DyConv), capture local features through dynamic convolution kernels, and improve the segmentation performance of Parkinson’s images. The experimental data used clinical data, and the improved algorithm achieved Dice, Precision, Recall, and mIou of 80.27%, 89.99%, 88.88%, and 82.38% in substantia nigra segmentation, respectively. The experimental results show that this algorithm can achieve more precise segmentation of Parkinson’s images.