Benefiting from the advantages of effective cost, contactless detection, and privacy preservation of WiFi sensing, WiFi sensing-based authentication studies are gaining more and more attention. However, most of the existing WiFi sensing-based authentication methods focus on identity recognition, lacking dedicated illegal intruder detection. Additionally, WiFi-based authentication methods directly extracts gait features from the channel state information (CSI) obtained from the reflection path, ignoring the significant changes in CSI when the user passes through the near-field line of sight (LOS) path. To this end, we propose a novel WiFi-based authentication method for narrow spatial environments, considering the fusion of LOS near- field features and reflection path features. We design a WiFi Feature Fusion Network (WFFN) based on feature fusion to identify legal users. WFFN synchronously extracts the dynamic features reflecting the pattern of individual walking and the near- field features reflecting the most significant variations caused by passing through the near-field LOS path. In addition, we design an Inception-based WiFi Convolutional Network (IWCN) to enhance the ability of WFFN to perceive the spatial characteristics of subcarriers.