In response to the inaccurate and inefficient search for static targets by autonomous underwater robots (AUV), a static target search method for the marine environment based on Bayesian Gaussian mixture model (BGM) is introduced. Firstly, the abnormal data point information transmitted by the underwater sensor network is processed using the Parzen window theory to establish a static target distribution probability map. Secondly, to improve the search efficiency of the AUV under time constraints, the BGM is employed to analyze the probability map of static target distribution to identify the regions with peak probabilities. Thirdly, Glasius bio-inspired neural network (GBNN) topology organization model is constructed to reflect the characteristics of the search area. The activity value of each neuron is updated by the propagation between adjacent neurons and the detection reward, and the intelligent agent plans the search path based on the distribution of GBNN's activity values. Finally, the future detection reward of the peak probability region is calculated and directly introduced into the external excitation input of GBNN neurons to solve the problem of poor global search due to propagation delay and attenuation. The GBNN-BGM and GBNN-GMM methods can achieve maximum cumulative detection rewards, with the GBNN-BGM method being more effective in task completion.