Here's what you've got, according to your circuit diagram:
Obviously I've just used a TL07x OPA because I don't have a model for the OPA551 handy. I've injected 0.1V of noise at 10 kHz into the power rail, and this is what you get as output with 100 mV input at 1kHz:
Here's the circuit reworked with a proper active virtual ground, and bypassing on the voltage divider that virtual ground is derived from:
There the second OPA could really be a TL07x or any other suitable OPA, it doesn't need to be a beast like the OPA551. And here's the output under the same conditions:
Noise gone! The issue was the impedance of the virtual ground, and adequate filtering on the source of that virtual ground.