Deep learning has made significant advancements in polarization imaging through scattering media. However, there remains a deficiency in effective analysis and control of the abundant information embedded in polarization images during training. Therefore, we propose a de-scattering framework with stepwise adjustment of polarization features (FSAPF) for high-performance imaging through scattering media. This framework implements a physically guided hierarchical learning, in which supervision processes are progressively refined from global structure to local polarization details. To directly embed polarization priors into feature representations, a polarization learning module (PLM) is introduced, which regulates feature interactions by enforcing physical consistency constraints, thereby enabling the FSAPF to learn robust representations. In addition, by directly embedding polarization priors into the dynamic loss mechanism, polarization features can be enhanced during training, which further enhances the FSAPF's generalized robustness in changing scenarios. We conduct a series of validation experiments to verify the validity and superiority of the FSAPF. The experimental results show that the FSAPF can perform significantly in target recovery tasks under different scattering environments. And comparative and ablation experiments can both achieve better results.

-methyladenine (6mA) is an important epigenetic modification that plays a vital role in various cellular processes. Accurate identification of the 6mA sites is fundamental to elucidate the biological functions and mechanisms of modification. However, experimental methods for detecting 6mA sites are high-priced and time-consuming. In this study, we propose a novel computational method, called Ense-i6mA, to predict 6mA sites. Firstly, five encoding schemes, i.e., one-hot encoding, gcContent, Z-Curve, K-mer nucleotide frequency, and K-mer nucleotide frequency with gap, are employed to extract DNA sequence features. Secondly, eXtreme gradient boosting coupled with recursive feature elimination is applied to remove noisy features for avoiding over-fitting, reducing computing time and complexity. Then, the best subset of features is fed into base-classifiers composed of Extra Trees, eXtreme Gradient Boosting, Light Gradient Boosting Machine, and Support Vector Machine. Finally, to minimize generalization errors, the prediction probabilities of the base-classifiers are aggregated by averaging for inferring the final 6mA sites results. We conduct experiments on two species, i.e., Arabidopsis thaliana and Drosophila melanogaster, to compare the performance of Ense-i6mA against the recent 6mA sites prediction methods. The experimental results demonstrate that the proposed Ense-i6mA achieves area under the receiver operating characteristic curve values of 0.967 and 0.968, accuracies of 91.4% and 92.0%, and Mathew's correlation coefficient values of 0.829 and 0.842 on two benchmark datasets, respectively, and outperforms several existing state-of-the-art methods.

Automatic roads detection is an essential task for traffic safety and intelligent transportation systems. Recently the long-wave infrared (LWIR) polarization imaging-based road detection technique has obtained significant progresses. However, the joint analysis among multiple polarization characteristics, sparse inter-channel information (along the$z$-axis), and dense intra-channel information (inside the$x$-$y$plane), have not been considered effectively, hindering the effective detection of many road areas. Additionally, most of the existing methods often encounter a challenging trade-off between achieving high precision and maintaining a lightweight design. To tackle these issues, this paper presents a novel Lightweight Multi-Pathway Collaborative 2D/3D Convolutional Networks (LMPC2D3DCNet) with a small number of parameters for full-time road detection. Our LMPC2D3DCNet is the first attempt to incorporate 2D and 3D convolutional networks to balance extraction for sparse inter-channel polarization information and dense intra-channel polarization information, in which a new Cross 2D-3D Non-Local Attention (C2D3DNLA) network is proposed to derive respective latent features by exploiting both local and global polarization correlations. Meanwhile, it also follows the design of a multipath network structure that elegantly fuses plenty of low-frequency, high-frequency, and multiscale polarization information, thus obtaining more accurate modeling for road regions. Extensive experiments on one public infrared polarization dataset of road scenes demonstrate that our proposed LMPC2D3DCNet (The code will release soon on https://github.com/XueqiangF) achieves PRE of 96.96%,RECof 96.71%,OAof 99.45%, F1 of 96.72, BER of 1.80% andIoUof 93.85%, and outperforms significantly state-of-the-art methods.