Research On The Segmentation Algorithm Of Seabed Mineral Images Based On Depth Features

With the continuous development of practical applications such as autonomous driving and vision systems,the importance of effective image processing has become increasingly prominent,and the requirements for image segmentation speed and segmentation accuracy have also been increasing.The pros and cons of image segmentation results have a fundamental impact on higher-level image processing tasks,such as image semantic understanding.Recent years,the insufficiency of traditional image segmentation methods such as spectral clustering and watershed has gradually emerged while the deep learning is more and more prevalent,and depth features have gradually become the main image segmentation feature extraction methods.Using depth features to describe textures,shapes,colors,etc.not only surpasses traditional methods in characterizing effects,but also has great potential in processing power.Therefore,the efficient use of depth features to solve computer vision problems is the focus of future research.This paper studies how to effectively segment seabed mineral images.The main work is as follows:(1)The application of traditional segmentation methods in seabed mineral images was explored,and the corresponding application system was developed based on the watershed algorithm.(2)The end-to-end segmentation algorithm of convolutional neural network based on encoder-decoder structure is studied,and the convolution feature maps of decoding structures with different scales are merged.By testing on public datasets,a network activation function suitable for seabed mineral image segmentation is selected,and the improved deep neural algorithm is compared with the traditional segmentation algorithm and the typical encoder-decoder structure to verify the effectiveness of the algorithm.Seabed mineral image processing based on digital image processing effectively combines computer application technology with mineral resource monitoring,which solves the problem of low efficiency and low accuracy based on manual methods.The research of this subject has certain theoretical significance and Realistic application value.