Holography is an emerging technology where phase-only spatial light modulator (SLM) is better than amplitude-only SLM because of its enhanced diffraction efficiency and rendering quality. However, phase-only interference pattern reconstruction is an ill-posed inverse problem, difficult to solve by standard supervised neural network approaches. Accurately representing the target input during reconstruction of phase-only hologram (POH) can become challenging given the inherent high spatial frequency of phase data and the complexity of its fringe patterns, expressing the inverse nature of the problem. Therefore, the model may be trained improperly, inducing major compromise in the predicted hologram’s quality. As a result, an unsupervised approach can be effective to resolve such a problem. This paper presents a novel combined U2-net-based computer-generated hologram (CUSNet-CGH) architecture, an end-to-end deep learning model comprising two distinctively designed combined convolution networks that accurately generate 2D POH. The proposed framework contains U2-net which is known for performing unsupervised learning tasks without relying on pre-trained backbones from labeled dataset, making it an ideal approach for conducting unsupervised training. The numerical reconstructions are compared with several other state-of-the-art methods such as holoencoder, holonet and CCNN-CGH to verify the enhanced quality. The average PSNR of the proposed end-to-end approach is as high as ~34 dB with an average SSIM of 0.91, featured by DIV2K dataset, outperforming the state-of-the-art methods analyzed in the paper. Additionally, the time required to reconstruct the predicted output of the proposed model is verified as competitive, challenging the latest benchmarks, thanks to its minimized number of trainable parameters, and ensuring that the model is computationally lightweight. Finally, experiments also demonstrate that the proposed model yields high-quality optical reconstruction.

Detecting fine, weak-textured defects with discontinuous boundaries on complex industrial surfaces is challenging due to interference from background textures and characters, as well as the scarcity of labeled data. To address this issue, we propose YOLO-SR, an engineering modification of YOLO11 tailored to defect segmentation on smart-card surfaces. Rather than introducing a new detection architecture, YOLO-SR reuses the backbone–neck–head design of YOLO11 and only adjusts a few modules to better capture elongated, low-contrast defects. The approach comprises two key components: first, embedding Strip Pooling (SP) within the C3K2 module to form C3K2_SP; second, a Rectangular Self-Calibration Module (RCM) is interposed after the top-level semantic layer. RCM generates rectangular gates to spatially recalibrate local responses, suppressing interference from complex textures and characters. To mitigate data scarcity and distributional bias, a texture-adaptive procedural defect synthesis strategy was developed. This strategy generates defect samples that conform to the background texture statistics of high-quality backgrounds. Experiments on the integrated circuit chip (ICChip) and signature plate (SignPlate) datasets show that YOLO-SR outperforms the YOLO11 baseline. Results indicate that SP and RCM complement each other by integrating directional priors from mid-to-high layers with top-level shape self-calibration. This enhances the visibility and localization stability of elongated defects while maintaining efficient inference.