海底物体搜寻、海底捞救、海底探宝、水雷探测以及智能水下机器人航行避碰等多方面都涉及海底小目标自动识别。将多波束前视成像声呐应用于海底小目标自动识别，图像的多区域、抗干扰、精细自动分割是关键是基础。海底小目标多波束前视声呐图像的多区域、抗干扰、精细自动分割是声呐图像处理领域的科学难题，项目研究旨在建立基于三维属性直方图的声呐图像精细自动分割方法，为这个科学难题提供一种有效的解决方案。项目拟推广申请人前期提出的一（二）维属性直方图概念，形成三维属性直方图的概念，给出三维属性直方图的定义；拟通过BP神经网络优化属性集参数，建立三维属性直方图的智能化构建方法；拟推广申请人前期提出的基于一（二）维属性直方图的分割方法，建立基于三维属性直方图的声呐图像精细自动分割方法。项目提出的三维属性直方图概念、基于BP神经网络的三维属性直方图智能化构建方法、基于三维属性直方图的声呐图像分割方法等均具原创性。

The automatic recognition of small targets lying on seabed is widely used in target searching, rescuing, treasure hunting, mine hunting, and navigating for autonomous vehicles under the sea. When a multi-beam forward-looking imaging sonar is employed for automatic recognition of small targets lying on seabed, with strong anti-interference ability, precisely and automatically segmenting a sonar image into multi-region is a crucial and underlying step. With strong anti-interference ability, precisely and automatically segmenting an image from a multi-beam forward-looking imaging sonar and small targets lying on seabed into multi-region is a difficult scientific problem in sonar image processing. This project aims to establish the methods for precise and automatic sonar image segmentation based on three-dimensional bound histograms in order to provide an effective solution to the above problem. The project intends to expand the concept of one- (two-) dimensional bound histograms proposed by the applicant and propose the concept of three-dimensional bound histograms and give the definition of three-dimensional bound histograms. The project intends to optimize the parameters in bound sets by means of the BP neural network and establish the intelligent method for construction of three-dimensional bound histograms. The project intends to expand the image segmentation methods based on one- (two-) dimensional bound histograms proposed by the applicant and establish the methods for precise and automatic sonar image segmentation based on three-dimensional bound histograms. In the project, the concept of three-dimensional bound histograms, the intelligent method for construction of three-dimensional bound histograms by means of the BP neural network, and methods for sonar image segmentation based on three-dimensional bound histograms are original.