Image dehazing is a common operation in autonomous driving, traffic monitoring and surveillance. Learning-based image dehazing has achieved excellent performance recently. However, it is nearly impossible to capture pairs of hazy/clean images from the real world to train an image dehazing network. Most of existing dehazing models that are learnt from synthetically generated hazy images generalize poorly on real-world hazy scenarios due to the obvious domain shift. To deal with this unpaired problem arisen by real-world hazy images, we present Cycle Spectral Normalized Soft likelihood estimation Patch Generative Adversarial Network (Cycle-SNSPGAN) for image dehazing. Cycle-SNSPGAN is an unsupervised dehazing framework to boost the generalization ability on real-world hazy images. To leverage unpaired samples of real-world hazy images without relying on their clean counterparts, we design an SN-Soft-Patch GAN and exploit a new cyclic self-perceptual loss which avoids using the ground-truth image to compute the perceptual similarity. Moreover, a significant color loss is adopted to brighten the dehazed images as human expects. Both visual and numerical results show clear improvements of the proposed Cycle-SNSPGAN over state-of-the-arts in terms of hazy-robustness and image detail recovery, with even only a small dataset training our Cycle-SNSPGAN. Code has been available at https://github.com/yz-wang/Cycle-SNSPGAN.

This paper studies a practically meaningful ship detection problem from synthetic aperture radar (SAR) images by neural network. We broadly extract different types of SAR image features and raise the intriguing question if these extracted features are beneficial to (1) suppress data variations (e.g., complex land-sea backgrounds, scattered noise) of real-world SAR images, and (2) enhance the features of ships that are small objects and have different aspect (length-width) ratios, therefore resulting in the improvement of ship detection. To answer this question, we propose a SAR-ship detection neural network (called 3SD-Net for short), by newly developing Bidirectional Coordinate Attention (BCA) and Multi-resolution Feature Fusion (MRF) and a center point distribution module (CPDM) based on CenterNet. In detail, we first develop BCA to make 3SD-Net focus on ship features as much as possible while ignoring the background noise. Second, we leverage MRF to enhance the spatial information of small-scale ships yet solve the non-trivial problem of small-scale and shallower pixels easily lost after deep convolution in SAR images. Moreover, considering the varying length-width ratio of arbitrary ships, we study the probability distribution around the ship center. We concentrate on enhancing the distribution function of the ship center, thereby significantly improving the performance of the basic CenterNet detector. This improvement is achieved without incurring additional computational and time costs. The experimental results obtained from the public SAR-Ship and SSDD datasets demonstrate the superior performance of our method compared to its competitors. Specifically, our 3SD-Net achieves AP values of 91.66% and 90.22% on the two datasets respectively, outperforming YOLOV7 (90.31% and 87.32%) and EfficientVit (90.06% and 90.08%). Source code will be released upon publication .

Background: This study aimed to develop machine learning models to predict Low Cardiac Output Syndrome (LCOS) in patients following cardiac surgery using machine learning algorithms.Methods: The clinical data of cardiac surgery patients in Nanjing First Hospital between June 2019 and November 2020 were retrospectively extracted from the electronic medical records. Six conventional machine learning algorithms, including logistic regression, support vector machine, decision tree, random forest, extreme gradient boosting and light gradient boosting machine, were employed to construct the LCOS predictive models with all predictive features (full models) and selected predictive features (reduced models). The discrimination of these models was evaluated by the area under the receiver operating characteristic curve (AUC) and the calibration of the models was assessed by the calibration curve. Shapley Additive explanation (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were used to interpret the predictive models.Results: Data from 1,585 patients [982 (62.0%) were male, aged 18 to 88, 212 (13.4%) with LCOS] were employed to train and validate the LCOS models. Among the full models, the RF model (AUC: 0.909, 95% CI: 0.875–0.943; Sensitivity: 0.849, 95% CI: 0.724–0.933; Specificity: 0.835, 95% CI: 0.796–0.869) and the XGB model (AUC: 0.897, 95% CI: 0.859–0.935; Sensitivity: 0.830, 95% CI: 0.702–0.919; Specificity: 0.809, 95% CI: 0.768–0.845) exhibited well predictive power for LCOS. Eleven predictive features including left ventricular ejection fraction (LVEF), first post-operative blood lactate (Lac), left ventricular diastolic diameter (LVDd), cumulative time of mean artery blood pressure (MABP) lower than 65 mmHg (MABP < 65 time), hypertension history, platelets level (PLT), age, blood creatinine (Cr), total area under curve above threshold central venous pressure (CVP) 12 mmHg and 16 mmHg, and blood loss during operation were used to build the reduced models. Among the reduced models, RF model (AUC: 0.895, 95% CI: 0.857–0.933; Sensitivity: 0.830, 95% CI: 0.702–0.919; Specificity: 0.806, 95% CI: 0.765–0.843) revealed the best performance. SHAP and LIME plot showed that LVEF, Lac, LVDd and MABP < 65 time significantly contributed to the prediction model.Conclusion: In this study, we successfully developed several machine learning models to predict LCOS after surgery, which may avail to risk stratification, early detection and management of LCOS after cardiac surgery.

Data in industrial informatics could be high-dimensional and mislabeled. The irrelevant or noisy features present significant challenges to high-dimensional mislabeled instances detection. Traditional methods often adopt the following method: firstly, searching for the relevant subspace; secondly, using the obtained feature subspace for mislabeled data detection. However, such a two-step method separates the relationship between feature selection and mislabeled detection, which can degrade the mislabeled detection performance. In this paper, we propose a novel sequential ensemble method SENF that aggregates the above two phases. Our method learns the sequential ensembles to obtain a refined feature subspace and improves detection accuracy by iterative sparse modeling with noise scores as the regression target attribute. Through extensive experiments on several real-world high-dimensional datasets, we show that SENF performs significantly better or at least similar to the individual baselines as well as the existing state-of-the-art label noise detection methods.