基于分形增强及孤立森林的声呐数据目标检测

Target detection for sonar image data based on isolation forest algorithm with fractal enhancement

针对现有声呐图像检测和判读目标精度不足的问题,提出一种基于分形维数增强及孤立森林算法的声呐数据人工目标检测算法。利用分形理论,首先计算原始声呐数据的位置及强度信号的分形特征,并将四维声呐数据提升至五维;然后利用增强后的五维数据构建孤立森林树型结构;最终采用遗传算法(genetic algorithm,GA)对孤立森林分离树的大小和数量进行自适应寻优,完成高精度的人工目标检测。基于真实海底地形声呐数据,所提算法能够同时准确检测多点虚拟人工目标,其准确率、模型评估指标(area under curve,AUC)等指标均优于现有方法。

Aiming at the problem of insufficient accuracy of existing sonar image detection and target interpretation,an artificial target detection algorithm was proposed for sonar data based on fractal dimension enhancement and isolation forest algorithm.Using fractal theory,the fractal characteristics of the original sonar data position and intensity signal was firstly calculated.And the sonar data was upgraded four to five dimension.Then,the enhanced five-dimensional data was used to construct an isolated forest tree structure.Finally,genetic algorithm(GA)was used to optimize the size and number of isolated forest trees adaptively to achieve high-precision artificial target detection.Based on the real seabed terrain sonar data,the proposed algorithm can detect multi-point virtual artificial targets accurately at the same time.And its accuracy,area under curve(AUC)and other indicators are better than the existing methods.