The task scheduling is a major research area in heterogeneous multi-core computing systems, in which managing complex tasks and achieving optimal efficiency presents challenges. Thus we propose a novel load balanced scheduling method based on multi-modal parallelism (Multi-modal Parallel Balanced Schedule, MPBS) by constructing a multi-modal parallel task computing model. The method integrates the improved greedy algorithm and Pierre Dellacherie algorithm to achieve dynamic scheduling of multi-modal tasks, and it is identified by iterative optimization search and fine-grained search compensation, which fully considers the utilization and load distribution for computing resources. Finally, the proposed method is implemented using hardware design and compared with several classic scheduling algorithms with respect to multiple performance metrics. The experimental results show that MPBS can obtain 35.4% decrease of makespan and 1.55x performance speedup, as well as 42.5% of load balancing rate reduction.