Accurate Recognition of Deep-Sea Small-Size Polymetallic Nodules Based on Multi-source Data and Deep Learning Model

Solid mineral resources are the fundamental material basis for maintaining the sustainable development of human society. The international seabed area contains vast and potentially valuable mineral resources, and deep-sea polymetallic nodules are one of the important ocean mineral resources. Taking the Peru Basin in the eastern Pacific Ocean as an example, this study aims to identify and classify small-scale polymetallic nodules occurred in the deep sea. Improving the resolution of deep-sea hydroacoustic images by utilizing super-resolution reconstruction methods. On this basis, the superpixel segmentation method is applied to construct a deep-sea object sample enhancement model, and the multi-dimensional heterogeneous features of the seabed objects are deeply explored to achieve effective construction of training samples. Under the constraint of geological seabed samples, an accurate seabed polymetallic nodule recognition model was thus established to achieve intelligent classification of seabed minerals based on multi-source data (including bathymetric data, backscatter data, etc.). Ultimately, by utilizing the model's generalization ability, the recognition and classification of untrained samples can be achieved, thereby advancing the application of the proposed algorithm in large-scale deep-sea mineral resource exploration.