A detection method for dense emitters based on a separation and boundary-aware collaborative enhancement detection network

To address the low detection efficiency of dense emitters in radio environment maps, this paper analyzes two fundamental challenges: feature confusion, which causes merged targets and increases missed detections, and boundary ambiguity, which blurs edges and raises false alarms. Together, they create a precision-recall dilemma. To this end, we propose a Separation and Boundary-Aware Collaborative Enhancement Detection Network, which contains two modules that respectively separate confused features and restore blurred boundaries. Experiments on a dense emitter dataset constructed from RadioMapSeer demonstrate that SBCE-Net achieves an F1 score of 0.988 alongside a recall of 0.991 and a precision of 0.985, outperforming seven existing methods in precision-recall balance. The FPPI-Recall curves show that this advantage holds across a broad range of confidence thresholds, and stratified density tests confirm that the method maintains the highest F1 at every density level, with a variation of only 0.116 from the densest to the sparsest condition. Ablation studies further verify that the two modules contribute complementarily, confirming the effectiveness of the collaborative design.