The first part computes the loss related to the predicted bounding box position coordinates  .  are the bounding box coordinates for ground truth data in training set.

 represents a constant used to give more penalty when there is a mistake. B is the number of bounding boxes. And  is the number of cells in the grid.

Similar equation is used to handle the bounding box width/height

Square root on the width and height in loss function equation is used to reflect that small deviations in small boxes matter more than large boxes. In general term this part of loss penalizes the bounding boxes with inaccurate height and width.