Light scattering is a pervasive problem in many areas. Recently, deep learning was implemented in speckle reconstruction. To better investigate the key feature extraction and generalization abilities of the networks for sparse pattern reconstruction, we develop the “one-to-all” self-attention armed convolutional neural network (SACNN). It can extract the local and global speckle properties of different types of sparse patterns, unseen glass diffusers, and untrained detection positions. We quantitatively analyzed the performance and generalization ability of the SACNN using scientific indicators and found that, compared with convolutional neural networks, the Pearson correlation coefficient, structural similarity measure, and Jaccard index for the validation datasets increased by more than 10% when SACNN was used. Moreover, SACNN is capable of reconstructing features 75 times beyond the memory effect range for a 120 grits diffuser. Our work paves the way to boost the field of view and depth of field for various sparse patterns with complex scatters, especially in deep tissue imaging.

Lipreading is a visual recognition of speech by using lip movement, which aims to recognise phrases and sentences spoken by a talking face without the audio. However, the existed models for lipreading suffer from slow training speed and insufficient performance. To accelerate the training speed of the model for lipreading, a batch group training algorithm is proposed, which groups all the data of different frames. In addition, a 3D‐MouthNet‐BLSTM‐CTC architecture for lipreading is proposed to improve model performance. It bases on a 3D convolutional neural network, MouthNet, two Bi‐LSTMs, and a CTC objective function. Experiment results in Oulu‐VS2 and self‐built dataset show that 96.2% accuracy rate is achieved on the Oulu‐VS2 dataset, and 93.8% accuracy rate is achieved on the GRID dataset. This article is about lipreading research. It mainly uses deep learning methods to study lip‐reading. A new network architecture and tests on public data sets are proposed to achieve the best results.

Local dimming technology focuses on improving the contrast ratio of the displayed images for a great visual perception. It consists of backlight extraction and pixel compensation. Considering a single existing backlight extraction algorithm can hardly adapt to images with diverse characteristics and rich details, we propose a stronger adaptive local dimming method with details preservation in this paper. This method, combining the advantages of some existing methods and introducing the combination of the subjective evaluation and the objective evaluation, obtains a stronger adaptation compared with others. Besides, to offset the luminance reduction caused in the backlight extraction process, we improve the bi-histogram equalization algorithm and propose a new pixel compensation method. To preserve image details, the Retinex theory is adopted to separate details. Experimental results demonstrate the effect of the proposed method on contrast ratio improvement and details preservation.

To address the intrinsic bottleneck coupling high sensitivity with a broad measurement range in optical fiber interferometric sensors caused by free spectral range (FSR) limitations, this paper proposes a novel intelligent demodulation method. This approach combines the optical Vernier effect with an attention-integrated deep convolutional neural network (DCNN-Attention). By connecting two parallel open-cavity MZIs with slightly disparate lengths, the Vernier effect yields an ultra-high refractive index (RI) responsivity reaching –152702.38 nm/RIU. However, conventional demodulation schemes constrain the detectable span to 0–9 mg/mL, yielding a sensitivity and R2 of –20.91 nm/(mg/mL) and 0.999151, respectively. To overcome this bottleneck, the DCNN-Attention model is introduced to perform end-to-end intelligent demodulation directly from raw interference spectra. This approach entirely eliminates the need for manual spectral preprocessing and Vernier envelope extraction, thereby avoiding the extraction errors and system noise impact associated with these conventional steps. Consequently, the sensing detection span is dramatically stretched to 0–200 mg/mL, representing a substantial 22-fold improvement. Experimental results demonstrate that within this broad range, the proposed model achieves a remarkable R2 of 0.999997, exhibiting superior demodulation accuracy. By synergistically combining physical sensing mechanisms with intelligent algorithms, this work effectively breaks the physical limitations imposed by the FSR. Ultimately, this method offers a new paradigm for high-performance optical sensing, showing broad application prospects in health monitoring and food safety analysis.