In the context of machine vision and industrial image processing, JPEG image enhancement remains a challenging task, especially under unknown compression parameters and severe visual artifacts. To address these issues, this paper proposes a robust and efficient enhancement method designed for resource-constrained visual systems. The framework adopts a cascaded architecture composed of multiple Enhancement Units (EUs), which progressively refine the visual quality of compressed images. A lightweight Quality Perception Module (QPM) is introduced to estimate a continuous representation of image quality, guiding each EU to perform quality-adaptive processing during both training and inference. Each EU is structured based on a U-Net design and consists of stacked Fast Blocks (FBs), while Context Modules (CMs) are embedded along the low-resolution path to model global semantic dependencies. In addition, a hybrid loss function is constructed by jointly considering spatial reconstruction loss, frequency domain consistency, and structural fidelity. Experimental results demonstrate that the proposed method outperforms existing techniques on multiple synthesized JPEG datasets, particularly in low-quality or double-compression scenarios. The method also achieves low computational cost, making it suitable for deployment in intelligent visual systems requiring real-time or near-real-time performance.