Spectrogram segmentation for underwater noise classification We use transfer learning on yolov7 to detect bounding boxes around potential dolphin whistles. Results on REPMUS datasets Data Type Training Labels Precision Recall mAP@0.5 REP23 430 0.59 0.50 0.31 REP23+aug 1300 0.66 0.69 0.55 REP23+21+aug 5000 0.79 0.69 0.77