Unsourced random access (URA) has garnered substantial interest for its applicability in massive machine-type communication (mMTC) scenarios. This paper introduces a novel multiple-antenna URA system that incorporates orthogonal frequency division multiplexing (OFDM) in multiple-input multiple-output (MIMO) systems, specifically designed for frequency-selective fading channels by exploiting the delay-angular domain sparsity. Each user's data is partitioned into two segments, one conveyed through selected pilots and the other through the payload. Crucially, we deploy the vector approximate message passing (VAMP) algorithm to first reconstruct users' wireless channels within the delay-angular domain. An iterative detector is then employed to decode the payload. Simulation results clearly demonstrate the proposed scheme's enhanced performance, significantly outperforming existing systems in managing frequency-selective fading channels.