How it works

The Sonar Noise Classifier is a trained Convolutional Neural Network optimized to identify noise in 3D point clouds generated by acoustic sensor platforms. All data goes through a pre- and post-processing stage to form the data into something the AI will understand: the data is rasterized into a high-resolution 3D voxel grid (zero-origin) before being transmitted to the Deep Learning Model for inference. Besides re-sampling the data into a format the AI can understand, this also has a side-benefit of completely anonymizing the data, as all spatial information is stripped during the voxelization process before transmission. When the classified voxels are returned from the AI, we post-process the data and map the results back to individual points to continue processing. The automation results in significantly less effort for the human processor allowing more energy to be focused on other, higher-value aspects of the production chain and, ultimately, increases the availability of data.