Weak target detection algorithm for passive radar based on deep learning

Aiming at the problem of the degradation of the detection performance of weak targets in the case of low signal-to-noise ratio, a weak target detection network model is proposed. This model converts one-dimensional radar data into two-dimensional image data and processes the images. A weak target detection network model for passive bistatic radar is put forward. The dataset is made by using the PPI (Plan Position Indicator) diagram. Based on the YOLOv8 basic framework, first of all, an attention mechanism module, namely the EffectiveSEModule, is added, which reduces the computational complexity and information loss while maintaining the performance of the model. Secondly, the neck network is replaced with the HSPAN-DySample network. With the help of a series of mechanisms such as its dynamic sampling, feature fusion, and detail enhancement, it can improve the detection ability for small targets. Then, the WIoU dynamic non-monotonic focusing mechanism is used to focus on anchor boxes of ordinary quality, thus improving the overall performance of the detector. Finally, an auxiliary detection head is added, which can better propagate the gradient, accelerate the training process, and prevent the problem of gradient vanishing. In order to verify the effectiveness of the algorithm, ablation experiments and comparative experiments are carried out. The experimental results show that the improved algorithm significantly reduces the situations of false detections and missed detections. The precision is increased to $90.5 \%$, an increase of $4.7 \%$, and the detection ability for weak targets is enhanced.