卷积神经网络的水下激光图像自动分割

Automatic segmentation of underwater laser image based on convolution neural network

研究卷积神经网络的水下激光图像自动分割方法,降低噪声对图像分割效果的干扰,促进海洋资源开发。深入研究水下激光图像成像原理,分析图像噪声来源,使用小波分析消除图像噪声,利用全卷积神经网络分割消噪后的图像,通过改造后的能量函数优化分割后图像的细节,并将优化结果保存,采用Gabor滤波器提取图像纹理特征,使用随机森林分类器完成图像分割优化与纹理特征的再分类,实现水下激光图像的精确分割。实验结果表明:选择0.3、0.5作为能量函数中区域项和新增能量项比例,可获得较理想的水下激光图像分割效果;能量函数改进后的收敛速度及泛化能力显著提升;该方法能很好地控制噪声对图像分割结果的干扰,水下激光图像自动分割效果优势显著。

The research of automatic segmentation method of underwater laser image of convolutional neural network is to reduce the interference of noise on image segmentation effect,and promote the development of marine resources.The principle of underwater laser image imaging is deeply studied,the source of image noise is analyzed,the wavelet analysis is used to eliminate image noise,and the full convolutional neural network is used to segment the denoised image.The details of the segmented image are optimized through the transformed energy function,and the optimal results are saved.The Gabor filter is used to extract the image texture features,and the random forest classifier is used to complete the image segmentation optimization and the reclassification of the texture features to realize accurate segmentation of the underwater laser image.The experimental results show that by choosing 0.3 and 0.5 as the ratios of the area term and the new energy term in the energy function,a more ideal underwater laser image segmentation effect can be obtained.The convergence speed and generalization ability of the improved energy function are significantly improved.This method can control the interference of noise on the image segmentation results,and the automatic segmentation effect of underwater laser images has significant advantages.