The Invertible Neural Network (INN) has recently gained prominence as a top-tier model in image steganography, enabling shared parameters between the concealment and revelation procedures. However, many existing INN-based steganography schemes face problems, including high memory requirements, extended training and inference times, and insufficient emphasis on high-frequency image regions. To address these challenges, we propose an enhanced lightweight image steganography scheme built on INN architecture with a progressive channel attention mechanism. This model employs multi-scale large-kernel convolution to efficiently grasp long-range feature correlations, thereby enabling it to acquire more comprehensive semantic information. Depthwise separable operations are incorporated into the large-kernel convolution to significantly reduce parameter counts and computational costs. Additionally, the progressive channel attention mechanism guides the model to focus on high-frequency areas of images during the training phase. The experiments carried out in the DIV2K dataset indicate that our method excels over others, delivering satisfactory results in terms of visual steganography quality, peak signal-to-noise ratio, and structural similarity metrics.