一种视听融合的水下目标识别方法研究

Study of underwater targets recognition based on audiovisual feature integration

特征提取是水下目标识别研究中最为关键的技术之一,特征参数的优劣将直接决定分类识别系统的性能。将声信号的听觉与视觉感知特征结合,应用于水下目标识别,通过实验得出如下结论,相比于单独应用听觉特征,融合特征的平均识别率能提高4%~6%以上,特别是将听觉特征与声谱图的Gabor小波变换特征、灰度-梯度共生特征进行融合后,分类性能较好,平均达到87%以上。

Feature extraction is one of the most important techniques of underwater targets recognition. Feature parameters directly determine the performances of classification systems. In this paper, a joint method which combines the auditory and visualized feature extraction methods is proposed and applied to underwater target recognition after feature selection and fusion. Experimental results show that fusion features achieve a better performance than a single audio feature, and the enhancement of recognition rate is 4%~6%, and that the fusion of audio feature with gray gradient co-occurrence matrix and Gabor small wave exchange feature can obtain an even better performance and the recognition rate is over 87%.