Advancing Quality and Detail: Enhanced-LapSRN for Chip Socket Image Super-Resolution

Recently, image super-resolution reconstruction models based on deep neural networks have achieved remarkable success. This paper introduces a network named Enhanced-LapSRN, designed using the classic CNN structure. This network not only effectively utilizes the pyramid structure from the LapSRN to construct high-resolution image residuals but also incorporates the architectural concept of DenseNet to facilitate feature reuse and prevent information loss. Additionally, we employ dilated convolution and convolution cascade techniques to expand the receptive field and introduce a convolution block attention module(CBAM) to focus more on high-frequency information in the image. This allows for the comprehensive integration of image features at different levels and results in comprehensive feature extraction. Further experiments confirm the strong performance of our method on public datasets, with a notable 0.49dB improvement in SET5, as well as an 8\% enhancement in positioning accuracy on our self-constructed chip socket datasets. Furthermore, our method is capable of real-time processing, ensuring high efficiency.