论文部分内容阅读
为提高高分辨率光学遥感图像港口自动检测的准确性,常需综合多类线索并进行复杂的特征提取、融合与分类推理,从而带来较高的计算复杂度。为此,仿生人类视觉注意机制,提出了一种复合线索视觉注意模型,综合利用高分辨率光学遥感图像港口多尺度底层特征和高层知识线索,实现了港口检测特征自然融合与综合分类推理。该方法在提高检测效果的同时较好地控制了计算量的增长,避免了复杂特征的大范围区域提取,采用多步快速算法降低了整个算法的计算复杂度,实现了计算资源受限条件下港口的快速定位与检测。同时,由于能将有限计算资源快速聚焦于最可能含有港口目标的区域,大大提高了目标检测方法响应的实时性。来自不同卫星的高分辨率光学遥感图像实验结果,验证了提出方法的有效性。
In order to improve the accuracy of automatic detection of high-resolution optical remote sensing image port, it is often necessary to synthesize multiple clues and perform complex feature extraction, fusion and classification reasoning, thereby resulting in higher computational complexity. For this reason, a visual attention model of compound cues is proposed by using biomimetic human visual attention mechanism. Based on the multi-scale bottom features and high-level knowledge clues of high-resolution optical remote sensing images, a natural fusion of port detection features and comprehensive classification reasoning are realized. This method not only improves the detection performance but also controls the growth of the computational complexity and avoids the extraction of large area of complex features. The multi-step fast algorithm reduces the computational complexity of the whole algorithm, Port rapid positioning and testing. At the same time, the ability to focus limited computational resources quickly on areas most likely to contain harbor targets greatly improves the real-time performance of target detection methods. The experimental results of high resolution optical remote sensing images from different satellites verify the effectiveness of the proposed method.