Intelligent Sensor Placement for Source Localization Based on Image Recognition of Hot Areas

In passive localization networks, the geometric configuration of the sensors (receivers) critically determines the source (emitter) localization accuracy. While most existing studies focus on optimizing sensor placement in ideal scenarios without placement constraints, this paper pays attention to address the practical challenge of irregular geographical terrain. The advanced image recognition technology is utilized to identify deployable areas from geographical images, thereby guiding practical sensor placement for passive localization. Specifically, we first propose an enhanced YOLOv11 model incorporating attention mechanisms and lightweight convolutions to accurately identify key geographical elements (KGEs). Second, to handle the complexity of sensor placement under irregular terrain constraints, we reformulate the problem as a sensor selection task and develop a modified Iterative Swapping Greedy (MISG) algorithm for efficient suboptimal solution. Simulations demonstrate that the proposed image recognition algorithm attains $87.4 \%$ accuracy in recognizing all KGE types, surpassing the original approach by $7.7 \%$, while the proposed placement algorithm achieves at least a $36.3 \%$ improvement in average regional localization accuracy compared to four baseline approaches.