Deep learning-based technologies have significantly improved the performance of image super-resolution networks, where lightweight convolutional neural networks (CNNs) are characterized by a
small number of parameters and low model complexity. However, their performance is greatly limited,
affecting the practical application of super-resolution methods. To address such issues, a lightweight
multi-scale large kernel convolution super-resolution reconstruction network is proposed, which fully
utilizes gating mechanisms and large kernel convolution. A multi-stage gated aggregation block is
introduced, using a series of gated units to obtain both global and local information; a multi-stage large
kernel attention block is proposed, which uses large kernel convolution to achieve a larger receptive field
and extract features from images at different scales; a focused gated aggregation block is introduced,
using skip connections to enhance the model's ability to reconstruct images. To validate the effectiveness
of this model, experiments were conducted on evaluation metrics such as the number of model parameters
and response speed, as well as subjective visual effects, and ablation analysis experiments were performed on each module of the model. The experiments show that compared to lightweight network
models in recent years, the proposed method achieves a significant improvement in peak signal-to-noise
ratio and structural similarity on four benchmark datasets while maintaining a similar or even reduced
number of parameters and response speed. This better balances model complexity and performance, and
achieves more effective subjective visual effects, leading to superior image reconstruction capabilities.
