论文部分内容阅读
针对目前显著性检测算法在复杂多目标遥感图像中检测能力不足问题，提出一种结合显著性检测和超像素分割的遥感信息提取算法。该算法首先通过graph-based visual saliency(GBVS)方法检测出原始影像中部分显著性较高的区域，然后利用simple linear iterative clustering(SLIC)方法分割显著区域，并修正显著区域边缘得到训练样本数据，进一步对训练样本进行统计学习，构造显著目标提取的阈值区间，最后实现对整幅超像素图像的显著目标提取。实验结果表明，该算法具有较高的准确率和召回率，能更加有效地检测出遥感图像中的显著目标，比目前主流的显著区域检测算法提取效果更好，可以很好地应用于具有明显显著区域的复杂多目标遥感图像信息提取中。
Aiming at the lack of detection ability of the saliency detection algorithm in complex multi-target remote sensing images, a remote sensing information extraction algorithm combining saliency detection and hyperpixel segmentation is proposed. The algorithm firstly detects some regions with significant saliency in the original image by the graph-based visual saliency (GBVS) method, and then divides the salient regions by the SLIC (simple linear iterative clustering) method, and fixes the salient regions to obtain training sample data. Further learning the training samples statistically, constructing the threshold range of significant target extraction, and finally achieving significant target extraction of the entire superpixel image. The experimental results show that the proposed algorithm has high accuracy and recall, and can detect salient targets in remote sensing images more effectively. Compared with the mainstream salient region detection algorithm, this algorithm has a better extraction effect and can be well applied to those with obvious Multi-target remote sensing image information extraction in salient areas.