Semantic Segmentation Network for Unstructured Rural Roads Based on Improved SPPM and Fused Multiscale Features DOI Creative Commons

Xinyu Cao,

Yongqiang Tian,

Zhixin Yao

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(19), P. 8739 - 8739

Published: Sept. 27, 2024

Semantic segmentation of rural roads presents unique challenges due to the unstructured nature these environments, including irregular road boundaries, mixed surfaces, and diverse obstacles. In this study, we propose an enhanced PP-LiteSeg model specifically designed for segmentation, incorporating a novel Strip Pooling Simple Pyramid Module (SP-SPPM) Bottleneck Unified Attention Fusion (B-UAFM). These modules improve model’s ability capture both global local features, addressing complexity roads. To validate effectiveness our model, constructed Rural Roads Dataset (RRD), which includes set scenes from different regions environmental conditions. Experimental results demonstrate that significantly outperforms baseline models such as UNet, BiSeNetv1, BiSeNetv2, achieving higher accuracy in terms mean intersection over union (MIoU), Kappa coefficient, Dice coefficient. Our approach enhances performance complex providing practical applications autonomous navigation, infrastructure maintenance, smart agriculture.

Language: Английский

Lightweight U-Net-Based Method for Estimating the Severity of Wheat Fusarium Head Blight DOI Creative Commons
Lei Shi, Zhihao Liu,

Chengkai Yang

et al.

Agriculture, Journal Year: 2024, Volume and Issue: 14(6), P. 938 - 938

Published: June 15, 2024

Wheat Fusarium head blight is one of the major diseases affecting yield and quality wheat. Accurate rapid estimation disease severity crucial for implementing disease-resistant breeding scientific management strategies. Traditional methods estimating are complex inefficient, often failing to provide accurate assessments under field conditions. Therefore, this paper proposes a method using lightweight U-Net model segmenting wheat spike spots estimate severity. Firstly, employs MobileNetv3 as its backbone feature extraction, significantly reducing number parameters computational demand, thus enhancing segmentation efficiency. Secondly, network has been augmented with Coordinate Attention (CA) module, which integrates lesion position information through channel attention aggregates features across two spatial dimensions. This allows capture long-range correlations maintain positional information, effectively while ensuring model’s efficient characteristics. Lastly, depthwise separable convolutions have introduced in decoder place standard convolutions, further parameter count maintaining performance. Experimental results show that Mean Intersection over Union (MIoU) reached 88.87%, surpassing by 3.49 percentage points, total only 4.52 M, one-sixth original model. The improved demonstrates capability segment individual conditions infestation, providing technical support identification research.

Language: Английский

Citations

2
GPS-free autonomous navigation in cluttered tree rows with deep semantic segmentation DOI Creative Commons
Alessandro Navone, Mauro Martini, Marco Ambrosio

et al.

Robotics and Autonomous Systems, Journal Year: 2024, Volume and Issue: 183, P. 104854 - 104854

Published: Nov. 8, 2024

Language: Английский

Citations

2
Semantic Segmentation Network for Unstructured Rural Roads Based on Improved SPPM and Fused Multiscale Features DOI Creative Commons

Xinyu Cao,

Yongqiang Tian,

Zhixin Yao

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(19), P. 8739 - 8739

Published: Sept. 27, 2024

Semantic segmentation of rural roads presents unique challenges due to the unstructured nature these environments, including irregular road boundaries, mixed surfaces, and diverse obstacles. In this study, we propose an enhanced PP-LiteSeg model specifically designed for segmentation, incorporating a novel Strip Pooling Simple Pyramid Module (SP-SPPM) Bottleneck Unified Attention Fusion (B-UAFM). These modules improve model’s ability capture both global local features, addressing complexity roads. To validate effectiveness our model, constructed Rural Roads Dataset (RRD), which includes set scenes from different regions environmental conditions. Experimental results demonstrate that significantly outperforms baseline models such as UNet, BiSeNetv1, BiSeNetv2, achieving higher accuracy in terms mean intersection over union (MIoU), Kappa coefficient, Dice coefficient. Our approach enhances performance complex providing practical applications autonomous navigation, infrastructure maintenance, smart agriculture.

Language: Английский

Citations

1