In light of the recent proliferation of unmanned aerial vehicles (UAVs) and the challenges posed by unauthorized flights interfering with air traffic, the precise detection and identification of UAV have become critical for ensuring security at low altitudes. This study introduces PAS-YOLO, an enhanced algorithm for detecting and recognizing UAV remote control signals, built upon the You Only Look Once version 12 (YOLOv12) framework, with the objective of improving UAV target identification capabilities. To augment the detection of small target signals and reduce the risk of remote control signal loss, a parallelized patch-aware attention (PPA) module is integrated into the backbone network. Addressing the limited feature representation capacity of YOLOv12, particularly the difficulty in distinguishing similar remote control signals through fine-grained features, the neck network is redesigned based on the Attentional Scale Sequence Fusion YOLO (ASF-YOLO) architecture. Furthermore, to broaden the receptive field and enhance the contextual extraction capability for signals with diverse time-frequency characteristics, the original area attention with C2f (A2C2f) module is refined by incorporating a switchable atrous convolution (SAConv) module. Experimental evaluations are performed using the publicly available Radio Frequency (RF) signal dataset DroneRFa, wherein the Short-Time Fourier Transform (STFT) is employed to generate a UAV time-frequency spectrum dataset. The results indicate that the proposed PAS-YOLO algorithm attains an average detection accuracy of 99.36% for mAP@50 and 75.38% for mAP@50:95 across 22 UAV remote control signal models. Compared to the baseline YOLOv12 model, these metrics represent improvements of 0.23% and 3.45%, respectively.