基于形状相似度的水下目标识别算法

在水下声纳图像目标检测与识别中,水下目标的识别问题是关键技术之一。基于形状相似度的概念,提出一种新的水下目标形状分类算法。首先对检测到的水下目标阴影区域进行规格化处理,采用改进的形状上下文方法对常见水下目标形状进行分类识别。与传统形状上下文方法相比,改进后的算法具有旋转不变性和模板自适应更新能力。通过对球形、圆柱形和圆台形三类目标的阴影区域进行仿真分类计算可知,该方法具有仿射不变性,分类准确率较高。

Mine target recognition based on shape similarity

In target detection and recognition from sonar images, recognition of mine target is a key technique. Based on the concept of shape similarity, a novel method for mine shape classification is proposed. Shallow region of mine is extracted and normalization. Modified shape context method is adopted to classify contour of mine targets. Compared with the conventional shape context method, the modified version has characteristics such as rotation invariance and template adaptive renewing. Simulation is performed to classify spheres, cylinders, and tapers, with a ratio of correct classification 92.7%. Results indicate that this method is affine-transform invariant and has high precision.