Deep-sea Sulfide Resource Evaluation Method Based On Deep Learning

Submarine hydrothermal sulfide is the product of volcanic activity and hydrothermal action.Because it is rich in a variety of metal elements,it is considered as a mineral resource that can be utilized by human beings in the future.Therefore,finding out the distribution of hydrothermal sulfide resources on the seafloor is a work of interest to scientists all over the world.Seabed video camera is an intuitive and effective technical means for polymetallic hydrothermal sulfide exploration.By identifying seafloor images,it is possible to identify lithology types,observe the seafloor environment,and find information on seafloor hydrothermal sulfide mineralization.The traditional manual image recognition method has low efficiency and high misjudgment rate,so this technology can only be used as an auxiliary method for submarine hydrothermal sulfide exploration.To this end,we introduce artificial intelligence technology(deep learning)to establish an automatic identification mode of seabed images to realize automatic classification and identification of seabed images,so as to improve the efficiency and accuracy of seabed video camera technology for exploration of seabed hydrothermal sulfides.Combining geological prior knowledge with knowledge gained from mid-Atlantic Ridge surveys.The seafloor lithology is divided into six types: pelagic sediments,ordinary basalts,pillow basalts,basalt breccias,transition zone sediments and hydrothermal sulfides.31,499 frames of seabed images were manually marked with the above six types of lithology and two types of invalid types to form a "training or verification atlas" for seabed image classification;Or verification data set" for training and verification to obtain a seabed image recognition model;use the recognition model to automatically identify seabed images for 27 video survey lines in the exploration area.Actual recognition effect:(1)The accuracy rate of Conv Ne Xt network training set is 99.3%,Res Net is 97.4%,Swin-Transformer is 97.3%;(2)Res Net validation set accuracy rate is 98.0%,Conv Ne Xt and Swin-Transformer are 97.9% %.It shows that deep learning has a high generalization ability to identify seafloor lithology and can be used for investigations in other exploration areas.(3)Deep learning only needs 0.018 seconds to recognize each frame of image,which is much faster than manual interpretation,and the frameby-frame recognition speed is higher than the video sampling speed(0.04 seconds/frame).Based on deep learning,it can efficiently analyze existing video data,but also can Realize the synchronization of seabed camera and lithology identification,quickly locate ore points,and realize intelligent investigation of seabed resources.Combined with artificial intelligence,the video survey line was identified second by second,and the geological mapping of the study area was completed.On this basis,five verification target areas were delineated,and the resource reserves were estimated by the block method,the average thickness of 10 m and 20 m ore bodies,and the area-tonnage model.The estimated resources in the study area were 38 Mt,25Mt,50 Mt and 26 Mt,respectively.It shows that the recognition technology of seabed video images based on artificial intelligence can greatly promote the ability of seabed hydrothermal sulfide exploration and resource evaluation.