An improved supervised and attention mechanism-based U-Net algorithm for retinal vessel segmentation

Computers in Biology and Medicine, Год журнала: 2023, Номер 168, С. 107770 - 107770

Опубликована: Дек. 2, 2023

Язык: Английский

---

Gpmb-yolo: a lightweight model for efficient blood cell detection in medical imaging

Health Information Science and Systems, Год журнала: 2024, Номер 12(1)

Опубликована: Март 11, 2024

Язык: Английский

---

Small and Dim Target Detection in IR Imagery: A Review

Опубликована: Янв. 1, 2024

Язык: Английский

---

Automatic Liver Tumor Segmentation from CT Images Using Graph Convolutional Network

Sensors, Год журнала: 2023, Номер 23(17), С. 7561 - 7561

Опубликована: Сен. 1, 2023

Segmenting the liver and tumors in computed tomography (CT) images is an important step toward quantifiable biomarkers for a computer-aided decision-making system precise medical diagnosis. Radiologists specialized physicians use CT to diagnose classify organs tumors. Because these have similar characteristics form, texture, light intensity values, other internal such as heart, spleen, stomach, kidneys confuse visual recognition of tumor division. Furthermore, identification time-consuming, complicated, error-prone, incorrect diagnosis segmentation can hurt patient's life. Many automatic semi-automatic methods based on machine learning algorithms recently been suggested organ segmentation. However, there are still difficulties due poor precision speed lack dependability. This paper presents novel deep learning-based technique segmenting identifying maps. Based LiTS17 database, comprises four Chebyshev graph convolution layers fully connected layer that accurately segment Thus, accuracy, Dice coefficient, mean IoU, sensitivity, precision, recall obtained proposed method according dataset around 99.1%, 91.1%, 90.8%, 99.4%, 91.2%, respectively. In addition, effectiveness was evaluated noisy environment, network could withstand wide range environmental signal-to-noise ratios (SNRs). at SNR = -4 dB, accuracy remained 90%. The model has satisfactory favorable results compared previous research. According positive results, expected be used assist radiologists specialist doctors near future.

Язык: Английский

---

Density-Aware U-Net for Unstructured Environment Dust Segmentation

IEEE Sensors Journal, Год журнала: 2024, Номер 24(6), С. 8210 - 8226

Опубликована: Янв. 31, 2024

Vision-based segmentation methods rely heavily on image quality, and mining environments are full of dust, which greatly reduces visibility. Efficient accurate dusty regions in can improve the performance unmanned vehicle environment perception. In this article, a dust method based novel density-aware nested U-structure convolutional neural network (DAUnet) is proposed. Compared with existing methods, our has three advantages. First, we introduce Residual channel-spatial attention (RCSA) block. The block contains two layers residual structure, extract features more efficiently. Second, difference expansion layer. This structure filters predicted probabilities, eliminates pixels lower then maps similar probability values to larger intervals, thus improving performance. Finally, for visualization, use probabilities reflect density, results smoother transition between edges background. addition, most current datasets generated by simulation tools, there lack open-source real-world datasets. Therefore, constructed MineDust dataset real open-pit environment. consists state images under different weather conditions complex scenes. Experiments demonstrate that algorithm achieve 79.64% mIoU, outperforms many methods.

Язык: Английский

---

DPANet: Dual Pooling‐aggregated Attention Network for fish segmentation

IET Computer Vision, Год журнала: 2021, Номер 16(1), С. 67 - 82

Опубликована: Сен. 10, 2021

Abstract The sustainable development of marine fisheries depends on the accurate measurement data fish stocks. Semantic segmentation methods based deep learning can be applied to automatically obtain masks in images data. However, general semantic cannot accurately segment objects underwater images. In this study, a Dual Pooling‐aggregated Attention Network (DPANet) adaptively capture long‐range dependencies through an efficient and computing‐friendly manner enhance feature representation improve performance is proposed. Specifically, novel pooling‐aggregate position attention module channel are designed aggregate contexts spatial dimension dimension, respectively. These two modules adopt pooling operations along information, respectively, thus reducing computational costs. these modules, maps generated by four different paths aggregated into one. authors conduct extensive experiments validate effectiveness DPANet achieve new state‐of‐the‐art well‐known image dataset DeepFish as well SUIM, achieving Mean IoU score 91.08% 85.39% while significantly FLOPs about 93%.

Язык: Английский

---

TiM-Net: Transformer in M-Net for Retinal Vessel Segmentation

Journal of Healthcare Engineering, Год журнала: 2022, Номер 2022, С. 1 - 17

Опубликована: Июль 11, 2022

retinal image is a crucial window for the clinical observation of cardiovascular, cerebrovascular, or other correlated diseases. Retinal vessel segmentation great benefit to diagnosis. Recently, convolutional neural network (CNN) has become dominant method in field, especially U-shaped CNN models. However, conventional encoder vulnerable noisy interference, and long-rang relationship fundus images not been fully utilized. In this paper, we propose novel model called Transformer M-Net (TiM-Net) based on M-Net, diverse attention mechanisms, weighted side output layers efficaciously perform segmentation. First, alleviate effects noise, dual-attention mechanism channel spatial designed. Then self-attention introduced into skip connection re-encode features long-range explicitly. Finally, SideOut layer proposed better utilization from each layer. Extensive experiments are conducted three public data sets show effectiveness robustness our TiM-Net compared with state-of-the-art baselines. Both quantitative qualitative results prove its practicality. Moreover, variants also achieve competitive performance, demonstrating scalability generalization ability. The code available at https://github.com/ZX-ECJTU/TiM-Net.

Язык: Английский

---

CI-UNet: melding convnext and cross-dimensional attention for robust medical image segmentation

Biomedical Engineering Letters, Год журнала: 2024, Номер 14(2), С. 341 - 353

Опубликована: Янв. 8, 2024

Язык: Английский

---

DMF-Net: A Dual Remote Sensing Image Fusion Network Based on Multiscale Convolutional Dense Connectivity With Performance Measure

IEEE Transactions on Instrumentation and Measurement, Год журнала: 2024, Номер 73, С. 1 - 15

Опубликована: Янв. 1, 2024

Pan-sharpening is a remote sensing image processing technique whose main objective to generate images with high spatial and spectral resolution by combining panoramic (PAN) low multispectral (MS) images, this improves the detail of remotely sensed while preserving rich information. Inspired other deep learning methods on panchromatic sharpening, we propose dual-stream fusion network (DMF-Net) based multi-scale convolution dense connectivity, aiming improve quality fused images. DMF-Net uses novel attentional mechanism (MSKA) as well connectivity extract key features in different resolutions, fuses using module (FB), finally generates high-resolution multispectral(HRMS) In our experiments, tested proposed method three datasets, QuickBird, WorldView-2, Maryland, variety evaluation metrics, achieved PSNR 36.4065, 38.7740, 31.1169, respectively, also ranked terms metrics. The experimental results show that has significant advantages over existing techniques quantitative qualitative evaluations, can effectively reduce distortion. Furthermore, comparing schemes, demonstrate robustness generalizability method.

Язык: Английский

---

Intelligent Inspection Guidance of Urethral Endoscopy Based on SLAM with Blood Vessel Attentional Features

Cognitive Computation, Год журнала: 2024, Номер 16(3), С. 1161 - 1175

Опубликована: Апрель 4, 2024

Язык: Английский

---