With the increasing complexity of communication environments, factors such as extended transmission distances, limited antenna gain and internal noise of array elements often place the received signal data under low Signal-to-Noise Ratio (SNR). In such scenarios, where the noise power far exceeds the signal power, the signal features are highly susceptible to being overwhelmed by noise, rendering them blurred and indistinguishable, which substantially reduces the separability between signal and noise and fundamentally undermines the performance of traditional algorithms. To address the challenge of feature ambiguity in Direction of Arrival (DOA) estimation under low SNR, we propose the Physical Logic Reconstruction and Weight Allocation Network (PLR-WANet). Specifically, we propose a Physical Logic Reconstruction (PLR) module to capture the intrinsic computational relationships between the real and imaginary parts of complex signals. A Channel Residual Connection (CRC) module is employed to enable dynamic weighting, thereby enhancing the extraction of key physical channel features. A Multi-Label Classification (MLC) module with nonlinear mapping is constructed to improve fine-grained angular resolution. Simulation results demonstrate that our method significantly improves the performance of the DOA estimation under low SNR compared to the existing methods.