Identification Of Seamount Slope Depositional Types Based On Deep Learning

Seamounts are first-order geomorphic forms in the deep sea and are of great research value.Cobalt-rich crusts,which are widely distributed on seamount slopes,are rich in a variety of key metals and are an important strategic resource.Slope topography controls the spatial distribution of sediment types such as seamount crusts within the slope,so it is an important task to study the spatial distribution pattern of sediment types on seamount slopes.The study implements the use of artificial intelligence method to discern the sedimentation type of seamount by video data.Geological station surveying is the most direct and accurate way in cobalt-rich crust resource exploration and resource assessment,although its high cost and low efficiency.Seabed video camera is the most visual,fast and effective tool in seabed resource investigation.By identifying seafloor video data,it is possible to observe seafloor activity,changes in seabed environment and understand seafloor sedimentation types.Manual recognition of seabed images is inefficient and difficult to ensure consistent recognition criteria.In this paper,deep learning method is used to identify seafloor images and solve the problem of identifying sedimentary types of seamount slopes.A new solution is proposed to solve the location problem of deep-sea towing video.In the processing of video data,the method of morphological comparison of a huge number of topographic profiles are used to match and compare the water depth profile of the camera survey line with tens of thousands of topographic profiles in the study area.And In the comparison result,the location of the topographic profile with the highest similarity to the bathymetric profile of the camera line is the location of the camera line,based on which the positioning and correction of the camera data is carried out,which methodologically solves the problem of positioning the video data when the ultra-short baseline(USBL)data is extremely abnormal.It also provides a solution to the problem of locating near-seabed sounding equipment and its sounding data.Deep learning network models were constructed to determine the sedimentary type of seamount slope by using camera data.First,In order to determine the sedimentary types of seamount,the "seamount image dataset" is constructed by selecting typical images from seamount camera videos.The images are corresponding to the sedimentary types of seamount slopes,and are labeled and divided into four sedimentary types: pelagic sediments,transition zone sediments,continuous crusts and discontinuous crusts.Deep learning methods were selected to build Swin-Transformer,Conv Ne Xt,Mobile Net and Res Net networks,and the seamount image dataset was used for training and validation to build a seamount slope image recognition model.And then video data recognition in the study area was carried out.The recognition results of these networks were 91.7% for the Res Net training set and92.9% for the validation set,77.2% for the Swin-Transformer training set and 78.5%for the validation set,86.2% for the Conv Ne Xt training set and 86.3% for the validation set,and 86.3% for the Mobile Net training set.86.3%;Mobile Net training set accuracy of 90.5%,validation set accuracy of 90.3%.The results show that the deep learning recognition of seamount slope sediment images has high generalisation ability and can be used for crust survey area work.Intelligent recognition of seamount slope video survey lines using the Res Net network with the highest accuracy takes only 0.021 seconds per image,and the frame-by-frame recognition speed is also higher than the video sampling speed,allowing efficient analysis of existing video data as well as identification of seabed sediment types at the same time as the camera,in parallel,to quickly locate mineral sites and achieve intelligence in the recognition of seabed resource survey visual data.The seamount sedimentation model was further established and validated,which is restricted by topographic.Comparison was made between the identification results and the slop-constrained seamount slope deposition law(Du et al.,2020),and correlation analysis was conducted with geological sampling.The results show that4.8° ± 1.2° is approximately the threshold range that controls pelagic sediments,transition zone sediments and cobalt-rich crusts on ME seamount slopes.Slope surfaces with slopes less than 4.8° are mainly covered by pelagic sediments;slope surfaces between 4.8° ± 1.2° are mainly covered by transition zone sediments;and cobalt-rich crusts are found on seamount slope surfaces with slopes greater than 4.8°,with seamount slope topography controlling the spatial distribution of all types of sediments.