Abstract: Countering enemy phased-array radars by mounting electronic warfare equipment on swarm platforms is one of the impor- tant tactical means to make up for the shortcomings of single-unit countermeasure capabilities. Before achieving a comprehensive identification and prediction of the working status and the intention of the phased-array radars within the group, and making corresponding interference decisions, the interaction and the intelligent comprehensive analysis of the information of identical radiation source grasped by individuals within the group is an indispensable link. Aiming at the challenges in existing methods, such as heavy network interaction burden, information redundancy, difficulty in reflecting radar behavior characteristics, and large knowledge base scale, when a cluster platform mounts electronic warfare equipment to counter enemy phased-array radars, a framework of decentralized swarm radar countermeasure system based on a combination of swarm intelligence and neural networks for information fusion is designed in the paper. Within this framework, the optimal subgroups characterizing radiation source features can be formed in a self-organization manner by means of an individual local interaction mechanism. A cascaded convolutional neural network method based on model migration is employed. Initially, each individual neural network classifier within the group is trained independently. Subsequently, in the course of inter-subgroup interaction, each individual acquires the network weights and biases of other individuals, and cascades them with its own network, thereby achieving transfer learning and comprehensively characterizing the features of radiation source. Finally, the feasibility of the proposed algorithm is validated by means of simulation.