Better approach for exempting background as false positive.

Hello Everyone,

I am working on a project for detecting football, so I have made a detector using SSD mobilenet v2 which almost detect perfectly but there are many false positive, which detects background object as football with high accuracy. So for reducing this problem I have gathered the images of that object retrain the model by adding those objects as background, for example, My model was detecting Clock as a ball, so I added some Images for Clock as background and retrained the model, the model then perfectly differentiate between Clock and a ball. My question is that is there a better Approach than this for reducing the false positive in the detector.

Please let me know your suggestions or improvements for the detector.

Thank you.