Fast semantic segmentation for remote sensing images with an improved Short-Term Dense-Connection (STDC) network

Mengjia Liua,Peng Liu,Lingjun Zhao,Yan Ma,Lajiao Chen,Mengzhen Xu

DOI: https://doi.org/10.1080/17538947.2024.2356122

IF: 4.606

2024-06-04

International Journal of Digital Earth

Abstract:It is hard to accomplish fast semantic segmentation on large remote sensing images, since current neural networks with numerous parameters often rely on significant computational resources. Our team proposes an improved fast semantic segmentation model based on short-term dense-connection network (RepSTDC). We introduce a structure reparameterization and coordinate attention into STDC networks. By structure reparameterization, we transform the multi-branch structure into a comparable single-branch configuration during the inference process. By replacing the traditional channel attention with a coordinate attention mechanism, we enhance the attention mechanism with considering channel relationships and long-distance position information, and then it saves the memory usages. We conducted thorough experiments to assess the efficacy of network components of RepSTDC on the several benchmark datasets. Additionally, we compared our proposed approach with state-of-the-art methods. Our RepSTDC model can well balance the accuracy performances, computing speed, and memory usage in most cases. It achieves fast segmentation by significantly reducing parameters but without obviously compromising performances compared to other methods.

geography, physical,remote sensing

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper aims to address the problem of rapid semantic segmentation of large-scale remote sensing images. Current deep learning-based neural networks typically require a large number of parameters and computational resources, which is a challenge in practical applications. The paper proposes an improved fast semantic segmentation model—RepSTDC (Reparameterized Short-Term Dense-Connection), which achieves efficient segmentation tasks through structural reparameterization and coordinate attention mechanisms.

#### Main Contributions:
1. **Structural Reparameterization**: Converts multi-branch structures into single-branch structures to simplify computational complexity during inference.
2. **Coordinate Attention Mechanism**: Introduces the Coordinate Attention (CA) mechanism to replace traditional channel attention mechanisms, enhancing the model's ability to capture channel relationships and long-distance positional information while saving memory usage.

#### Experimental Results:
Experiments show that the RepSTDC model performs well on multiple benchmark datasets, significantly reducing the number of parameters while maintaining high accuracy, thereby improving segmentation speed and reducing memory consumption. Compared to existing state-of-the-art methods, RepSTDC achieves a better balance in most cases.