Haze degrades image clarity, contrast, and fine details, causing perceptual ambiguity. Image dehazing aims to restore visibility, but remains challenging in noisy and complex environments. To address these challenges, prior-based dehazing techniques have been widely explored for effective transmission estimation and scene radiance recovery. The Dark Channel Prior (DCP) uses the minimum intensity across all color channels to estimate transmission in dense haze regions efficiently. This approach works well in non-sky regions with rich textures and darker intensity. However, DCP may not work well in sky regions. On the other hand, the Bright Channel Prior (BCP) uses high-intensity information of pixels to achieve efficient dehazing in bright regions, such as light-dominant regions in the atmosphere, which covers the sky. However, BCP may not work well in non-sky regions. Most existing dehazing algorithms use either DCP or BCP to efficiently recover scene radiance. This results in reduced performance in different regions. To overcome this limitation, a region-based fusion approach is developed to achieve efficient scene radiance recovery from DCP and BCP. The fused image is enhanced using a Retinex-based image decomposition technique, which consists of gamma correction of illumination images and wavelet-based denoising of reflectance images. The performance of dehazing depends on the tuning parameters of the above-mentioned techniques. Also, the ground truth may not be available in real-time scenarios. Therefore, a no-reference (blind) adaptive parameter selection scheme is developed to handle the absence of ground truth in real-world images. The most suitable parameters for DCP and BCP are determined using a grid search guided by the NIQE metric, ensuring improved perceptual quality without requiring reference images. The method is evaluated on I-Haze, O-Haze, Dense-Haze, NH-Haze, and Reside (Reside-6K, RTTS, and UAHI) datasets using SSIM, PSNR, NIQE, and BRISQUE metrics. Experimental results demonstrate consistent improvement across diverse haze conditions. Moreover, the proposed framework operates without any training requirements, removing the necessity for annotated datasets and model optimization processes, all while ensuring low computational complexity, with an average processing time per image is just 0.423 s. Furthermore, the proposed approach demonstrates comparable performance to state-of-the-art methods and is validated through real-world object detection tasks. The source code of the proposed method is publicly available at: RegionBasedAdaptiveHybridImageDehazing.

Accurate and timely identification of surface defects in steel strips plays a critical role in maintaining product quality. The process of inspecting steel strips, however, has historically been a manual, time-consuming task dependent on experienced personnel and technicians. With modern Automated Visual Inspection (AVI) systems the process has become faster, but there are still limitations. To further improve the inspection process there is a need for companies to find ways of working with AVI-systems.In this study we have designed a Convolutional Neural Network (CNN) for a Swedish steel strip producer with the aim to replicate the results of an AVI-system installed in their production line. The designed CNN achieves an overall accuracy of 98% on the provided dataset. Next step is to test the model on larger datasets with more varied acquisition conditions to further assess generalization and identify potential weaknesses not visible in the current testing setup.

In this paper, we propose a nondestructive automated approach for classifying stacked roundwood into frozen and non-frozen phase state. More specifically, ultrawideband (UWB) radio transmission data was used to train a neural network. For this approach, data from 175 spruce logs stacks was acquired at different temperatures. Three features were extracted based on the radio transmission signals' zero crossings and amplitude information. These features were fed to a feed-forward neural network for binary classification purposes. A classification accuracy of 84.9% was achieved. These preliminary results show that time-domain features acquired from UWB radio transmission data can be used for classification of roundwood in frozen and non-frozen phase state.

Attention deficit hyperactivity disorder (ADHD) is a prevalent neurodevelopmental disorder characterized by inattention, impulsivity, and hyperactivity. Although serious games (SGs) are effective for ADHD assessment and intervention, integrating deep learning with data enhancement remains limited. In this study, we develop an automated framework for ADHD diagnosis using SG-based electroencephalography (EEG) frequency-band data. We employ a multiscale temporal convolutional network (MSTCN) and a conditional generative adversarial network (cGAN) for training-time augmentation. Experiments are conducted on the ADHD FOCUS dataset under keyboard-controlled and EMOTIV-controlled sessions, with signals segmented into 2 -second and 6 -second windows. Performance is evaluated using Leave-One-Subject-Out (LOSO) cross-validation, where the cGAN is trained only on training subjects and augments only the training set. Using real data, accuracies of 94.95% (keyboard) and 93.24% (EMOTIV) were achieved, with 3× cGAN augmentation, keyboard accuracy improved to 98.52% while EMOTIV remains comparable.

Feature extraction reduces high-dimensional image data into compact and informative descriptors, improving classification efficiency and performance. In this study, moisture content classification is performed using color image analysis. RGB channels are first separated, and 26 features are extracted from each channel, including statistical measures (mean, standard deviation, variance, median, minimum, maximum, skewness, kurtosis, Shannon entropy, and Renyi entropy) and texture features derived from Gray-Level Co-occurrence Matrix (GLCM). GLCM features contrast, correlation, energy, and homogeneity—are computed across four orientations (0°, 45°, 90°, and 135°). To enhance performance, Chi-square-based feature selection is applied to rank and select the most discriminative features. A Support Vector Machine (SVM) with a radial basis function (RBF) kernel is used for classification. Feature-level fusion of RGB channel combinations is also investigated. The model is evaluated using 4-fold and 10-fold cross-validation. The proposed method achieved a highest classification accuracy of 91.54 and 90.42 for 3-class and 5-class classification of moisture content, respectively. Results demonstrate that optimal feature selection and channel combinations significantly improve classification accuracy and robustness.

This paper addresses workload allocation for parallel machines subject to stochastic degradation in order to improve maintenance synchronization. In such systems, maintenance actions are often performed simultaneously across machines, which creates a trade off between premature maintenance and unexpected failures. Existing workload allocation strategies mainly focus on reducing failure probability or balancing degradation but do not explicitly aim at synchronizing maintenance conditions.

To address this issue, this paper proposes a framework combining stochastic degradation modeling and workload optimization. Machine degradation is modeled using a Gamma process in which the degradation rate depends on the assigned workload. A Sequential Quadratic Programming optimization is used to dynamically allocate workloads in order to maximize the probability that all machines reach the maintenance window defined by preventive and failure thresholds at the same time. Remaining Useful Life estimates are used to trigger workload reallocations and maintenance decisions.

Monte Carlo simulations compare the proposed strategy with a deterioration based workload allocation method from the literature. The results show that the proposed approach maintains a low proportion of corrective maintenance while achieving a high proportion of preventive maintenance with fewer workload reallocations.

This paper reports an approach to classify incipient faults in power transformers using deep learning models. Dissolved fault gases have been used as key indices to understand catastrophic failures of transformers. A two-dimensional feature matrix based on five primary features extracted from fault gases has been computed. The number of features varied from 5 to 40 by performing linear and non-linear operations on the primary features. These are further converted into 2D images, and several deep learning models are applied for classification trials on the generated 2D images. Additive noise augmentation is adapted to increase the training data to reduce overfitting and bias. Weighted cross-entropy loss is employed to address the class imbalance. Experiments are performed on two field-validated datasets one of which is the IEC TC 10 dataset, and the other is the field data of 4580 transformer dissolved gas analysis (DGA) samples. The proposed approach is reported with 100% and 96.67% accuracy using the IEC TC 10 dataset and field dataset, respectively, during fault identification tests.

This paper presents an image-based framework for classifying fluid flow regimes into low and high-speed states by utilizing spatially localized texture features combined with machine learning techniques. Traditional approaches, such as Computational Fluid Dynamics (CFD) and Direct Numerical Simulations (DNS), often require extensive post-processing to extract fluid flow properties. This makes them time-consuming and less practical for real-time applications. To address this, the proposed method leverages the Local Binary Pattern (LBP) feature extraction technique. LBP effectively captures local neighborhood patterns and converts complex flow behaviors into quantifiable texture features from images of CFD. These features are then classified using various machine learning models, namely Random Forest (RF), Support Vector Machines (SVM), and k-Nearest Neighbors (k-NN). The LBP-based approach demonstrates excellent performance, with the k-NN classifier achieving a maximum accuracy of 0.9879 in the case of flow past an elliptical cylinder. Similarly, the SVM classifier attains up to 0.9540 accuracy for the flow past an airfoil. Evaluations cover a range of Reynolds numbers from 200 to 5000 and turbulence intensities of 5% and 20%, confirming the robustness and effectiveness of the method. A comparative analysis with other texture-based techniques, namely Local Ternary Pattern (LTP) and Gray Level Co-occurrence Matrix (GLCM), further highlights the advantages of the proposed method. The LBP approach outperforms LTP and GLCM by 14.5% and 2.4%, respectively, in terms of prediction accuracy. This demonstrates the superior capability of LBP in flow regime classification. The dataset used in this study is publicly available at: https://www.kaggle.com/datasets/palanithanarajk/fluid-flow-images

Anomaly Detection in Argon Oxygen Decarburization (AOD) Processes is crucial for optimal functioning and for uninterrupted production in steel industry. In this prestudy, we aim to detect an anomaly called calibration error on a flow sensor in AOD converter. To this purpose, we acquire data from Alleima, a steel company in Sweden. This data consists of signals from five different sensors including two flow sensors for measuring oxygen and argon flows, control signals for the corresponding valves and the pressure after the gases are mixed. This data is manually annotated as calibration error on a flow sensor-based anomaly (CEFSBA) and normal class by the experts. From each of the sensor signals, two features namely entropy and variance are extracted. These features are fed to classifier to classify a feature vector into normal and CEFSBA. Our proposed approach detected CEFSBA with an accuracy of 92.1% when using neural networks. Our prestudy, indicates that machine learning-based approaches can be used to identify CEFSBA in the AOD converter. However, in order to use in industry, the performance needs to be improved before deploying it for real time CEFSBA detection.

Accurately identifying moisture content in wood chips is crucial for optimizing energy production. This paper presents an automated classification approach using ultrawideband (UWB) radio transmission data. We collect data from 1,923 samples across four power plants and extract seven key features based on zero-crossings and amplitude information. To enhance classification performance, we apply Chi-square feature selection to identify the most significant features. These selected features are then fed into a classifier to determine moisture content levels. Our approach achieves a classification accuracy of 85.26%, demonstrating the effectiveness of the extracted features and the proposed methodology.