水下图像背景噪声的分类识别新方法

A new method for classification and recognition of background noise in underwater images

为对水下图像背景噪声进行分类识别,提出一种水下图像背景噪声的分类识别方法。首先,采用Kendall等级相关系数检验方法提取水下图像的均匀区域,剔除包含目标的图像块,建立数据集用于分类器训练和测试;然后,设计了一个轻量级的卷积神经网络,训练参数量比现有的分类网络大为缩小;最后,利用数据集对卷积神经网络进行训练和测试。实验结果表明,所提方法准确率达到98%以上,超过了传统的支持向量机方法。

In order to classify and recognitze the background noise of underwater image, a classification and recognition method of underwater image background noise is presented. Firstly, the homogeneous image regions are extracted based on the Kendall coefficient hypothesis testing method, and image blocks containing targets are deleted from the homogeneous image regions, and then, a data set for training and testing the classifier is established. Secondly, a lightweight Convolutional Neural Network is designed, and its training parameters are much smaller than the existing classification Convolutional Neural Network. Thirdly, the data set is used to train and test the Convolutional Neural Network. The results show that the accuracy is over 98%, which is better than the result of the traditional support vector machine method.