We reprocessed and interpreted Chang’E-4 Lunar Penetrating Radar (LPR) data collected until 14th February 2023, exploiting a new Deep Learning-based algorithm to automatically extract reflectors from a processed radar dataset. The results are in terms of horizon probability and have been interpreted by integrating signal attribute analysis with orbital imagery. The approach provides more objective results by minimizing the subjectivity of data interpretation allowing to link radar reflectors to their geological context and surface structures. For the first time, we imaged dipping layers and at least 20 shallow buried crateriform structures within the regolith using LPR data. We further recognized four deeper structures similar to craters, locating ejecta deposits related to a crater rim crossed by the rover path and visible in satellite image data.