河流是遥感图像中一种非常典型且重要的地理目标,对河流的自动检测在水资源调查及水利规划等方面有重大意义。在此提出了一种基于多特征融合和软投票方法的河流目标检测算法,该算法首先将图像分割成胞元,提取胞元的局部熵、纹理、光谱和颜色等特征,利用随机森林训练和分类,并利用基于形态学运算结合多判据投票法优化机器学习的粗检测结果,对优化后的粗检测结果利用水平集活动轮廓逼近河岸线。实验表明,该算法检测效果良好,对测试集的检测准确率达97.44%,在复杂背景下可以有效检测出河流。

River is a very typical and important geographical target in remote sensing images. The automatic detection of rivers is of great significance in water resources investigation and water conservancy planning. In this paper, a river target detection algorithm based on multi-feature fusion and soft voting method is proposed. The algorithm firstly divides the images into cells and then extracts the local entropy, texture, spectrum, and color feature of the cells. Random forest is used to train and classify. To optimize the rough detection result of machine learning, the morphology operation and the multi-criteria voting method is introduced. For optimized rough detection result, the level set active contour is used to approach the river shoreline. Experiments show that the proposed algorithm has a good detection effect, and the detection accuracy rate of test set reaches 97.44%. In addition, the river can be effectively detected in the complex background.