The performance of many multi-sensor systems can be significantly improved by using a priori environmental information and sensor data to plan the movements of sensor platforms that are later deployed with the purpose of improving the quality of the final detection and classification results. However, existing path planning algorithms and ad-hoc data processing (e.g., fusion) techniques do not allow for the systematic treatment of multiple and heterogeneous sensors and their platforms. This paper presents a method that combines Bayesian network inference with probabilistic roadmap (PRM) planners to utilize the information obtained by different sensors and their level of uncertainty. The uncertainty of prior sensed information is represented by entropy values obtained from the Bayesian network (BN) models of the respective sensor measurement processes. The PRM algorithm is modified to utilize the entropy distribution in optimizing the path of posterior sensor platforms that have the following objectives: (1) improve the quality of the sensed information, i.e., through fusion, (2) minimize the distance traveled by the platforms, and (3) avoid obstacles. This so-called Probabilistic Deployment (PD) method is applied to a demining system comprised of ground-penetrating radars (GPR), electromagnetic (EMI), and infrared sensors (IR) installed on ground platforms, to detect and classify buried mines. Numerical simulations show that PD is more efficient than path planning techniques that do not utilize a priori information, such as complete coverage, random coverage method, or PRM methods that do not utilize Bayesian inference.