Blind image quality assessment (BIQA) is a crucial technique for selecting high quality images, which are vital for reliable diagnosis, effective algorithms, and ensuring an optimal visual experience for users. Due to full-reference and reduced-reference IQA methods require reference images, the demand for no-reference (NR-IQA) techniques is more pronounced in practical scenarios. In recent years, NR-IQA models based on Convolutional Neural Networks have achieved significant success. However, existing methods treat image regions uniformly, contradicting the trend of the human visual system to prioritize attention to salient areas. Inspired by the Human Visual System (HVS), we propose the hypothesis that different regions contribute differently to perceived image sharpness, with significant regions exerting a greater influence on quality assessment. To validate this hypothesis, we conducted subjective experiments, revealing that applying the same degree of blur to foreground and background regions results in different quality degradations. Based on this observation, we introduce A Saliency-Aware NR-IQA Method by Fusing Distortion Class Information(SADCIQA). A hierarchical saliency aware module is designed to obtain weighted fusion of features from different regions of the image. Additionally, we employ a self-supervised method to train a quality aware module to extract image quality features. SADCIQA emphasizes salient regions while reducing the contribution of background areas, aligning quality predictions with subjective judgments. We conducted experiments on six datasets, demonstrating that our approach achieved state-of-the-art performance in both synthetic and real distortions.