Though the oceans cover the most parts of the earth's surface, the deepest ones which are called hadal zones are still very mysterious. It is a frontier area to explore the hadal zone using modern technology. Multi-beam sonar is one of the most common ways to get the information from the hadal zone. Due to complex and volatile environment in the ocean, such as water speed, ocean depth, topography and so on, multi-beam sonar image may exist noise and deficiency. In this work, polynomial interpolation which is a machine learning method was utilized to restore the multi-beam sonar image. Color information is one of the most important for multi-beam sonar image, especially for the description of submarine topography, thus in this work, a strategy for color image restoration was utilized. The experimental result shows that the proposed method can recover the most parts of multi-beam image.