Negative samples are so large

when I’m training yolov3, I find that negative samples are extremely larger than positive samples so that mAP is very low. How can I solve this problem?