Research On Mineral Image Segmentation Algorithm For Seabed Mineral Resource Assessmen

Deep-sea mineral resources,as important strategic resources,include oil and natural gas,phosphorite,hydrothermal sulfides,cobalt-rich crusts,polymetallic nodules,and future new energy natural gas hydrates.The development and utilization of these resources can provide important energy and material supplies,and are of great significance for promoting economic development and achieving sustainable development goals.According to the preliminary assessment of global mineral wealth,the value of seabed mineral resources reaches trillions of US dollars,among which polymetallic nodules and other surface mineral deposits have huge potential.However,the complexity of the deep-sea environment and the uneven distribution of resources pose great challenges to resource exploration and development.The use of computer vision and deep learning technology can effectively improve the accuracy and speed of image processing and analysis,thus completing the assessment of deep-sea mineral images.This will help accelerate the exploration and development of polymetallic nodules and other surface mineral deposits in our country,and enhance the core competitiveness of our marine sector.The development of computer vision technology enables computers to recognize and process complex digital images,thus achieving high-precision identification of deep-sea mineral resources.Combined with deep learning technology,the accuracy and efficiency of image recognition can be further improved.Deep learning models have great potential in processing image data.Through model training,automated segmentation and recognition of deep-sea mineral images can be achieved.With the continuous advancement of computer vision and deep learning technologies,their applications in deep-sea mineral resource exploration and development may become more widespread in the future,providing strong support for safeguarding national maritime rights and interests.Based on existing domestic and international research,this paper mainly focuses on the following work:First,the Mask R-CNN algorithm is used to address the issue of ineffective deep-sea mineral image segmentation.The effectiveness of this method is demonstrated by comparing it with U-Net,MU-Net,and CGAN.An evaluation system for deep-sea mineral images is developed based on the Mask R-CNN model.It supports image segmentation,morphological fitting of mineral particles,and prediction of mineral coverage,particle size,and abundance,providing statistical parameters for mining engineering.Secondly,for the problem of degradation of nodule mineral image quality,an improved Sea-thru method is proposed.It can effectively simulate the degradation model of deep-sea mineral image quality and enhance the degraded deep-sea images.This method can reduce the impact of insufficient and uneven lighting in the underwater imaging environment,reduce image distortion,and improve the accuracy of subsequent segmentation tasks.Finally,focusing on the problem that Mask R-CNN fails to solve the problem that small nodule particles are difficult to be detected and segmented,a deep-sea mineral image segmentation algorithm combined with object detection is proposed.It uses a coarse-to-fine segmentation approach,first performing a coarse-grained segmentation on the original image,and then detecting small nodules that are easily missed and performing a fine-grained secondary segmentation.The fusion of coarsegrained and fine-grained segmentation results improves the segmentation accuracy of existing deep-sea mineral images.