Abstract: To address redundant exploration, trajectory oscillations, and limited coordination efficiency in multi-robot multi-target search within unknown environments, we propose an improved Glasius bio-inspired neural network (IGBNN) and a distributed cooperative search framework. Firstly, we discretize the unknown workspace into grid cells that are one-to-one mapped to neurons, and update the external stimulus online based on the exploration map to form a dynamic neural activity landscape. Secondly, we introduce a Gaussian direction weighting derived from the turning angle, together with an activity transmission threshold, to suppress abrupt activity changes caused by discrete decisions, thereby reducing ineffective turns and alleviating local deadlocks. Finally, we design an environment-mediated distributed coordination mechanism for local information sharing and conflict avoidance among robots. Matlab simulations in a 40×40 complex scenario with 40 targets show that, while maintaining a 100% success rate, the proposed method reduces the average number of turns by 25.8% and the average number of steps required to complete the full-target search by 15.1% compared with GBNN, confirming its advantages in search efficiency and cooperative performance in complex unknown environments.