Ensuring the structural integrity of aircraft blades is critical to flight safety, yet conventional inspection methods remain heavily reliant on manual labor and are prone to oversight—especially when detecting subtle or minute surface defects. Recent advances in deep learning offer promising automation potential, but existing models often exhibit poor robustness and limited generalization across complex, real-world scenarios in aerospace environments. In this paper, we present ABS-Net, a novel multi-attribute recognition framework tailored for high-precision aircraft blade inspection. Built on the Mask2Former backbone, ABS-Net introduces a novel learnable query-channel attention module, which dynamically prioritizes defect-relevant semantic cues and enhances feature discrimination for barely visible anomalies. In contrast to traditional segmentation methods, ABS-Net goes beyond pixel-level detection by jointly estimating critical defect attributes—including size, color, and aspect ratio—enabling more informed maintenance decisions aligned with aviation safety standards. While comparative evaluation with baseline methods is conducted on the ADE20K benchmark, we further provide quantitative and qualitative results of ABS-Net on the Aircraft Blade Defect (ABD) dataset, demonstrating its practical effectiveness in real-world aerospace inspection scenarios.