A novel biologically-inspired target detection method based on saliency analysis for synthetic aperture radar (SAR) imagery

Fei Ma,Fei Gao,Jun Wang,Amir Hussain,Huiyu Zhou

DOI: https://doi.org/10.1016/j.neucom.2019.12.009

IF: 6

2020-08-01

Neurocomputing

Abstract:<p>Saliency Object Detection (SOD) models driven by the biologically-inspired Focus of Attention (FOA) mechanism can obtain highly accurate saliency maps. However, their application in the high-resolution Synthetic Aperture Radar (SAR) images faces some intractable problems due to complex background. In this paper, we propose a novel hierarchical self-diffusion saliency (HSDS) method for detecting vehicle targets in large scale SAR images. To reduce the influence of cluttered returns in saliency analysis, we learn a weight vector from the training set to capture the optimal initial saliency of the superpixels during saliency diffusion. Considering the background objects have multiple sizes, the saliency analysis is implemented in multi-scale space, and a saliency fusion strategy then is employed to integrate the multi-scale saliency maps. Simulation experiments demonstrate that our proposed method with these improvements can achieve more accurate detection and lead to less false alarms, compared to benchmark approaches.</p>

computer science, artificial intelligence

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the problem of target detection in high-resolution Synthetic Aperture Radar (SAR) images. Specifically, the paper proposes a novel Hierarchical Self-Diffusion Saliency (HSDS) method for detecting vehicle targets in large-scale SAR images.

#### Main Issues:
1. **Complex Background Interference**: Traditional SOD models face the problem of complex background interference when processing SAR images.
2. **Speckle Noise Impact**: Strong speckle noise in SAR images interferes with the saliency initialization of superpixels, thereby affecting the results of saliency detection.
3. **Multi-Scale Object Detection**: The size of targets (such as vehicles, ships, buildings, etc.) in SAR images varies greatly, requiring multi-scale spatial analysis.

#### Solutions:
1. **Initial Saliency Optimization**: Optimize the initial saliency of superpixels by learning a weight vector from the training set, thereby reducing the impact of speckle noise.
2. **Multi-Scale Fusion Strategy**: Perform saliency analysis in multi-scale spaces and use hierarchical fusion rules to integrate multi-scale saliency maps, improving detection accuracy.
3. **Edge Clue Detection**: Use edge clues to quickly locate suspicious target areas, reducing background interference.

Through these improvements, the method proposed in the paper can reduce the false alarm rate while maintaining high detection accuracy. Experimental results show that this method outperforms existing benchmark methods.