A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Bridge component segmentation for health monitoring an enhanced DeepLabV3+ model with lightweight network and multi-scale channel attention mechanism
https://orcid.org/0000-0002-3042-4512
Due to the influence of various factors, such as complex environments and sustained load effects, the long-term service life of bridge structures will lead to a gradual deterioration in performance. Therefore, bridge health monitoring is of utmost importance, and component identification is a crucial step in evaluating the overall structural integrity of bridges. With the advancement of deep learning algorithms, semantic segmentation methods can effectively classify and identify bridge components in complex environments, thereby facilitating the assessment of their state. Nevertheless, the conventional methods for segmenting bridge components suffer from drawbacks such as intensive computation, inadequate feature extraction, and low segmentation accuracy, failing to meet the requirements of current bridge health monitoring. Consequently, this paper proposes a bridge component segmentation method based on an improved DeepLabV3 + model, named the DeepLabV3-MS, which is based on an enhanced DeepLabV3 + model. This method utilizes MobileNetV2 as the backbone network to reduce parameter count and improve the computational speed of the model. The Strip Pooling (SP) is also integrated into ASPP, known as SP_ASPP, to enhance the capture of more comprehensive contextual information. Additionally, the Multi-scale Channel Attention Mechanism (MS-CAM) is incorporated to enhance the integration efficiency of multi-semantic and multi-scale features. The results indicate that compared with the original DeepLabV3 + model, the Mean Intersection over Union and Mean Pixel Accuracy of the DeeplabV3-MS model increased by 5.90%, and 4.92%, respectively. Furthermore, in comparison to the classic models PSPNet and U-Net, DeeplabV3-MS demonstrated an increase of 19.50% and 8.88% in MIoU and MPA, respectively, as well as 13.50% and 5.34%, respectively. The proposed method has demonstrated superior performance across various evaluation metrics, exerting a significant impact on the health monitoring and safety assessment of bridge components. Furthermore, it offers valuable technical support for research and applications in related fields.
Bridge component segmentation for health monitoring an enhanced DeepLabV3+ model with lightweight network and multi-scale channel attention mechanism
https://orcid.org/0000-0002-3042-4512
Due to the influence of various factors, such as complex environments and sustained load effects, the long-term service life of bridge structures will lead to a gradual deterioration in performance. Therefore, bridge health monitoring is of utmost importance, and component identification is a crucial step in evaluating the overall structural integrity of bridges. With the advancement of deep learning algorithms, semantic segmentation methods can effectively classify and identify bridge components in complex environments, thereby facilitating the assessment of their state. Nevertheless, the conventional methods for segmenting bridge components suffer from drawbacks such as intensive computation, inadequate feature extraction, and low segmentation accuracy, failing to meet the requirements of current bridge health monitoring. Consequently, this paper proposes a bridge component segmentation method based on an improved DeepLabV3 + model, named the DeepLabV3-MS, which is based on an enhanced DeepLabV3 + model. This method utilizes MobileNetV2 as the backbone network to reduce parameter count and improve the computational speed of the model. The Strip Pooling (SP) is also integrated into ASPP, known as SP_ASPP, to enhance the capture of more comprehensive contextual information. Additionally, the Multi-scale Channel Attention Mechanism (MS-CAM) is incorporated to enhance the integration efficiency of multi-semantic and multi-scale features. The results indicate that compared with the original DeepLabV3 + model, the Mean Intersection over Union and Mean Pixel Accuracy of the DeeplabV3-MS model increased by 5.90%, and 4.92%, respectively. Furthermore, in comparison to the classic models PSPNet and U-Net, DeeplabV3-MS demonstrated an increase of 19.50% and 8.88% in MIoU and MPA, respectively, as well as 13.50% and 5.34%, respectively. The proposed method has demonstrated superior performance across various evaluation metrics, exerting a significant impact on the health monitoring and safety assessment of bridge components. Furthermore, it offers valuable technical support for research and applications in related fields.
Bridge component segmentation for health monitoring an enhanced DeepLabV3+ model with lightweight network and multi-scale channel attention mechanism
https://orcid.org/0000-0002-3042-4512
Due to the influence of various factors, such as complex environments and sustained load effects, the long-term service life of bridge structures will lead to a gradual deterioration in performance. Therefore, bridge health monitoring is of utmost importance, and component identification is a crucial step in evaluating the overall structural integrity of bridges. With the advancement of deep learning algorithms, semantic segmentation methods can effectively classify and identify bridge components in complex environments, thereby facilitating the assessment of their state. Nevertheless, the conventional methods for segmenting bridge components suffer from drawbacks such as intensive computation, inadequate feature extraction, and low segmentation accuracy, failing to meet the requirements of current bridge health monitoring. Consequently, this paper proposes a bridge component segmentation method based on an improved DeepLabV3 + model, named the DeepLabV3-MS, which is based on an enhanced DeepLabV3 + model. This method utilizes MobileNetV2 as the backbone network to reduce parameter count and improve the computational speed of the model. The Strip Pooling (SP) is also integrated into ASPP, known as SP_ASPP, to enhance the capture of more comprehensive contextual information. Additionally, the Multi-scale Channel Attention Mechanism (MS-CAM) is incorporated to enhance the integration efficiency of multi-semantic and multi-scale features. The results indicate that compared with the original DeepLabV3 + model, the Mean Intersection over Union and Mean Pixel Accuracy of the DeeplabV3-MS model increased by 5.90%, and 4.92%, respectively. Furthermore, in comparison to the classic models PSPNet and U-Net, DeeplabV3-MS demonstrated an increase of 19.50% and 8.88% in MIoU and MPA, respectively, as well as 13.50% and 5.34%, respectively. The proposed method has demonstrated superior performance across various evaluation metrics, exerting a significant impact on the health monitoring and safety assessment of bridge components. Furthermore, it offers valuable technical support for research and applications in related fields.
Bridge component segmentation for health monitoring an enhanced DeepLabV3+ model with lightweight network and multi-scale channel attention mechanism
Jiang, Tianyong (author) / Huang, Yali (author) / Hu, Chunjun (author) / Li, Lingyun (author)
Advances in Structural Engineering ; 28 ; 939-951
2025-04-01
Article (Journal)
Electronic Resource
English
Semantic Segmentation of Cracks Using DeepLabv3+
| Springer Verlag | 2023
Segmentation of Concrete Surface Cracks Using DeeplabV3+ Architecture
| Springer Verlag | 2023
Image Segmentation Method for Sweetgum Leaf Spots Based on an Improved DeeplabV3+ Network
| DOAJ | 2022
| DOAJ | 2024
Sweetgum Leaf Spot Image Segmentation and Grading Detection Based on an Improved DeeplabV3+ Network
| DOAJ | 2023